Intelligent video analysis is a hotspot in current security technology application, but some factors restrict the popularization and application of security intelligent video analysis technology. Therefore, this paper specifically analyzes the deep application bottlenecks of security intelligent video analysis technology and proposes corresponding countermeasures.
With the promotion of safe city construction, the application of security video surveillance has become increasingly popular. With the rapid expansion of the security video surveillance system, traditional manual monitoring has been difficult to meet the actual needs of security in terms of abnormal goals, behavior, event recognition efficiency, accuracy, real-time and so on. The advanced concepts and innovative application modes of security intelligent video analytics application technology are refreshing. However, after nearly ten years of development, there is still a huge contrast between its actual application results and user expectations, and it has almost fallen into the deadlock of “a thousand functions of product features, poor depth of application, and chaotic market supervisionâ€. How to break through the bottleneck of the application of intelligent video analytics for security, requires market surveillance, product development, system integration, and professional users to work together.
I. Application bottlenecks and countermeasures of intelligent video analysis technology Brief introduction of security intelligent video analysis technology
Security intelligent video analysis technology originated in computer vision technology. It comprehensively applies techniques such as image enhancement processing. Based on the separation of goals and backgrounds, the target characteristics information is compared with pre-set templates or rules to automatically identify interest. The target, behavior, event, or data generates an alarm, and if necessary, the abnormal target can be automatically tracked and linked with other security facilities, significantly improving the overall efficiency and capability of security prevention and control. At the same time, the use of security intelligent video analysis technology in massive historical video information to achieve intelligent query search based on time, location and semantic characterization, can provide efficient verification means for the reconnaissance office.
Basic architecture
The overall architecture of the security video analytics application system mainly includes the following three modes:
Front-end embedded deployment mode: The intelligent analysis function module is placed in front of a video capture front-end device, and a shared camera image processing chip or dedicated high-performance DSP is used as an intelligent analysis engine to realize intelligent analysis of abnormal targets, behaviors, and events. This kind of architecture is not affected by the image transmission link, which eases the pressure of video transmission and background storage. At the same time, intelligent analysis using uncompressed original images has good real-time performance, strong pertinence, and ease of deployment on a large scale. The disadvantage is that it is difficult to support complex intelligent analysis applications, and due to the pressure of front-end processing, higher requirements are placed on the integrated heat-dissipation process and stability of the front-end device, and it is relatively easy to be constrained by environmental factors. In addition, the front-end embedded intelligent analysis software is difficult to upgrade, and to some extent, the difficulty and cost of installation, commissioning, operation and maintenance are increased.
Background server deployment mode: Intelligent analysis applications are implemented through dedicated servers (groups) or dedicated clients deployed in the security monitoring center. This architecture effectively solves the inconsistency of the front-end processing capabilities. It can implement complex target, behavior, and event intelligence analysis through powerful background processing capabilities; facilitates integration with other professional application systems and expands the scope of smart analysis applications; intelligent analysis engine background running environment Guaranteed, system work stability is greatly improved, and it is convenient for intelligent analysis software upgrade, effectively reducing the pressure of system operation and maintenance. In short, this architecture is suitable for phased fixed-scenario intelligent analysis applications (such as large-scale event security), but it requires high performance for back-end server configuration. The level of intelligent analysis is easily limited by the bandwidth of image transmission links, and real-time performance is not ideal. Large-scale smart analytics applications can't be done.
Distributed hybrid deployment mode: Place some relatively simple video pre-processing functions on the camera (or video server) to complete the extraction, packaging, and sending of target feature data. The background intelligent analysis engine is responsible for receiving target feature data, completing and presetting rules. Or modules analyze and compare and automatically generate alarms. This architecture reduces the pressure on the front-end intelligent analysis. The back-end server only needs to analyze the target data, and it is convenient to integrate with other professional application systems. It improves the efficiency and capability of intelligent analysis and is more suitable for large-scale, high-performance intelligent analysis applications. However, the system The overall cost of deployment and operational maintenance pressure cannot be underestimated.
basic skills
Abnormal target recognition tracking. Identify, lock and track abnormal dynamic targets that occur in a specific period of time and in a specific area, and automatically record their movement trajectories.
Abnormal events, behavior identification alarm. Identify abnormal events or behaviors such as cross-border, legacy, stranded, stranded, fall, trail, fight, steal, and gather within a specified area and automatically report an alarm. Feature data acquisition. License plate recognition, face recognition, traffic flow monitoring, and real-time monitoring of crowdedness in public areas.
Massive video smart search. According to the semantic description of the target, behavior, and event features, and the auxiliary information such as location and time of the comprehensive event, the video information needed in the massive historical video information is automatically searched to improve the efficiency of the investigation of the staff of the department's investigators.
Second, security intelligent video analysis technology application status
At present, the application of security intelligent video analysis technology mainly focuses on urban public security management (such as public security bayonet information identification, crossing vehicle inspection and control, Internet cafe face comparison, etc.), and large-scale event security (such as the Shanghai World Expo campus entrance and exit portrait collection and comparison, virtual Waterfront electronic fence, etc.), management of the prison and other business areas. The comparatively successful application model is based on feature data acquisition comparisons. At the same time, in the surveillance sites where monitoring scenes are relatively fixed and lighting conditions are better, the intelligence of the objects, events such as illegal cross-border, overtime detention, confrontation, abnormal aggregation, etc. is supervised. Analyzing the alarm application has achieved certain results.
As far as the overall application level is concerned, due to factors such as the effectiveness of intelligent analysis algorithms, the complexity of the background, and the uncertainty of lighting conditions, the intelligent video analysis application of abnormal targets, behaviors, and events in complex open environments is still more challenging than the practical requirements for distance security. Big gap. The license plate recognition rate of the security bayonet and city road crossing is relatively high (more than 90%), but there is still much room for improvement based on the depth application of license plate recognition. Static portrait matching technology can achieve more than 90% accuracy in comparisons when the quality of the comparison source and comparison target is relatively high (such as portrait photos provided by second-generation residents), and it has higher actual combat applications. Value, but the real-time comparison technology for face collection in the open environment is limited by many factors such as site lighting conditions, “blacklist†quality, and the degree of cooperation, and the accuracy ratio is about 60-70%. Currently, it can only be deployed as a large-scale event in real time. Aids.
Third, the main factors restricting the application of intelligent video analytics security
The lack of illumination at night in the open environment, bad weather (fog, snow, rain, dust, etc.), image compression processing, and the limited bandwidth of the network transmission link have caused image quality to decline, which has brought inherent difficulties to security intelligent video analysis; Complex anomalous behaviors and event modeling are difficult, and the corresponding intelligent analysis algorithms do not have high recognition performance. Targets are too close to the background or the background is cluttered, which leads to difficulties in target segmentation and feature information extraction. The moving target is blocked by the background and multiple moving targets obstruct each other. Target feature information is incomplete. The above factors are likely to cause adverse consequences such as false alarms, missed alarms, and tracking difficulties, which will severely restrict the actual combat performance improvement of the security intelligent video analysis application system.
The efficiency of the embedded intelligent video analysis algorithm based on D1 or HD resolution is not high, leading to the shortage of front-end DSP hardware resources, lack of front-end equipment integration, and lack of heat-dissipating technology to further limit the performance of DSP. The hardware processing capability and stability of the embedded intelligent analysis engine have been improved. It has become another major factor that restricts the application level of security intelligent video analytics.
Due to the fact that most of domestic security intelligent video analysis application products are from foreign manufacturers, and there is no scale application effect, the product pricing is generally abnormally high; manual calibration, background software debugging and other workloads during the implementation of the project are large, and some high-end product debugging even The need for direct participation of foreign manufacturers and technical personnel, coupled with the relatively high failure rate of embedded front-end equipment, makes it difficult to implement remote maintenance such as fault diagnosis and software upgrades, leading to high operating and maintenance costs for application systems. The above factors result in the low cost performance of security intelligent video analysis applications, which severely limits the effective promotion of security intelligent video analysis applications.
Mechanism level
At present, the construction of the security video surveillance system is still at an extensive stage of scale expansion. As the main unit of construction, security users are still paying more attention to the "monitoring coverage rate," and the degree of attention to prior warning is not enough. The direction of intelligent video analysis is The investment in R&D funding is also obviously insufficient.
Security intelligent video analysis product development lacks effective mechanisms. On the one hand, most security users do not seriously sort out application requirements. On the other hand, the product R&D department lacks an in-depth understanding of the security industry, resulting in intelligent video analysis products that have a thousand functions and lack of targeting. Sex. At the same time, most system integrators do not give full play to the bridge between product R&D and users. Some integrators even exaggerate the functions of intelligent video analysis products and seriously damage the market image of intelligent video analysis products.
In addition, from the perspective of market supervision, due to the lack of authoritative standards certification system, users can not effectively measure and control the product performance, and seriously affect the user's confidence in the application of security intelligent video analysis.
Fourth, the main countermeasures to promote security intelligent video analysis applications
Focus on the support of local R&D forces and concentrate on key technologies.
In order to cope with the current situation in which security and intelligent video analytics applications rely heavily on foreign technology, and at the same time effectively stop the inefficient research and development of simple imitation and repeated investment, it is necessary to carefully organize local professional research institutes to focus on basic and common key technologies to break through security. Intelligent video analysis depth application creates conditions.
Efficient intelligent video analysis algorithm: Improved target segmentation, feature extraction, target recognition, and dynamic tracking algorithms based on complex scenes (such as varying lighting conditions, closeness of target and background colors, background clutter, targets blocked by background, and changes in monitored field of view) Performance improves the capacity and effectiveness of static object libraries and dynamic rule bases, reduces the occupation of hardware computing resources, and improves the accuracy and real-time performance of automated analysis of complex targets, behaviors, and events for open, complex, and large-scale monitoring areas. At the same time, the overall structure of the intelligent analysis software is increased to effectively take into account the real-time and robustness of the intelligent video analysis software, and meet the intelligent analysis requirements for different resolution video streams such as SDTV and HD, making the intelligent analysis software architecture more rational and efficient; Based on the automatic calibration technology of target characteristics, the ability to adaptive "learning" of back images is enhanced, the probability of false alarms is reduced, and complicated manual calibration is avoided, the difficulty of system debugging is reduced, and the maintainability of the system is improved.
Image acquisition and comparison technology: Further improve the algorithm for image acquisition and comparison based on open space and no-match conditions, eliminate the influence of unfavorable conditions such as complex and changeable on-site lighting and free movement of personnel, improve the accuracy of real-time comparison of the system, and improve In fact, the application value of warfare is used to improve the availability of backstage portrait storage data by automatically eliminating algorithms such as redundant portrait data; and to improve the reliability and real-time performance of automatic search and query comparison of static heterosexual portrait targets based on large-capacity portraits. In addition, we will increase the research and development of portable and transportable portrait collections compared to terminals, increase suspicious target maneuver control capabilities based on portrait collection and comparison technologies, and vigorously expand the application of portrait acquisition and comparison technologies in the practical application.
Massive video intelligent search technology: further improve the performance of the video semantic analysis (including the semantic feature extraction of video, multi-granular event representation, etc.) algorithm performance, and improve massive video intelligent search based on target or event semantic description and event time, place, region, etc. Real-time and accuracy, while providing mobile target trajectory analysis and display; accelerate the development of portable intelligent video search and comparison of terminal equipment, facilitate the search and investigation of departmental reconnaissance personnel on the independent storage of historical video search, improve the efficiency of video forensics.
Blurring image processing technology: Based on the selection of low-lightness, high-definition cameras and bandwidth of the transmission network as much as possible, further increase the development of fuzzy image processing technologies based on image enhancement, image restoration, and super-resolution reconstruction to improve the weather. The blurry image processing capabilities caused by the target motion, the defocusing of the optical lens, etc. lay the foundation for the subsequent application of intelligent video analysis. Increase demonstration application efforts to improve the level of intelligent video analytics applications in actual combat.
The key to the breakthrough of security intelligent video analytics level lies in application. Only through actual combat can system performance be truly verified and continuously improved. As a security department, we can neither regard intelligent video analysis technology as an omnipotent "magic bullet", and we should not adopt a passive and passive attitude. Instead, we should actively and steadily improve security based on the application strategy of "demand traction and demonstration first." Video analytics application level. According to the development status of the security intelligent video analysis technology, combined with the urgency of security business requirements, priority can be given to the following demonstration applications:
First consider the occasions where the monitoring scene is relatively fixed, the lighting conditions are easily secured, and the target and behavioral information are highly identifiable (such as inside large-scale event security areas, indoor ATM machines, and prisons, detention centers, detention centers, interrogation rooms, etc.) Carry out automatic analysis of alarming applications of abnormal targets, behaviors and events to improve the safety management and control capabilities of closed environmental areas; develops and implements massive data back-end mining and intelligent association analysis applications such as public security bayonet information identification, electronic*, and city-level crossings, vehicle and vehicle inspection and control, etc. Security check and control capabilities of security bayonet and city road crossings; promotion of large-scale event security zone entrances and exits, railway station high-speed trains (EMUs) passenger self-entry passages, metro station underground pedestrian passages, and other area portrait collection and comparison demonstration applications. The personnel in the public agglomeration area have suspicious target deployment capabilities.
Secondly, intelligent target analysis based on complex backgrounds or scenarios, behavior, and event intelligence analysis are the difficulties in the application of security intelligent video analytics, and it is also a key area for security application. Therefore, under the premise of effective breakthroughs in the development of key technologies such as efficient intelligent video analysis algorithms, it is necessary to selectively select intelligent video analysis depth applications in areas with high crowd concentration and complex public security situations in order to enable intelligent video analytics applications. We will make substantial progress as early as possible in the critical areas of security operations and create necessary conditions for large-scale application of security intelligent video analytics.
V. Actively explore and practice, strive to break through the traditional mechanism constraints
Innovative product development mechanism. The performance of intelligent video analysis products is highly related to the characteristics of industrial applications. Considering the particularity and sensitivity of the security industry, if the traditional market independent R&D model is used, it is difficult for professional R&D institutions to accurately understand and grasp the complex business needs of security departments. It is also difficult to truly meet the security requirements. In view of this, it is necessary to innovate the “triangulation, integration, and user†tripartite cooperation mechanism in the form of a “joint key laboratory for smart video analytics application technology†based on the principles of “benefit sharing and risk sharingâ€. In the cooperative R&D process, security users mainly actively involved in the product R&D process by combing application requirements, organization and demonstration construction, and coordinated performance improvement; professional R&D institutions are working hard to improve intelligent video analysis algorithms based on a thorough understanding of security business requirements. Performance improves real-time performance and reliability while reducing the probability of false alarms and false negatives in the system. At the same time, it improves the cost-effectiveness of intelligent video analysis products. System integrators maximize their performance through camera selection, supplemental lighting, monitoring angle adjustment, and calibration. Provide high-quality surveillance images, highlight monitoring target feature information, and create basic conditions for intelligent video analytics applications. Through the above-mentioned cooperation mechanism, it promotes three-way benign interaction and leads the right direction of security video analytics R&D and application. More importantly, through the sharing of intellectual property rights and research results by security users, the cost of system construction can be effectively reduced, and the inherent enthusiasm, initiative, and creativity of security users can be fully mobilized to accelerate the promotion and application of security video analytics applications.
Expand funding support channels. At present, smart city construction based on core supporting technologies such as the Internet of Things has entered a substantial start-up phase in many domestic cities. Smart security is one of the key applications for the demonstration of Internet of Things, and it has also been included in the scope of smart city construction. Therefore, in order to effectively solve the problem of high price for security intelligent video analysis products due to insufficient application scale, it is possible to build “government scientific research funding support through security users, R&D institutions, and integrators jointly applying for key technology research and demonstration application projects. The company's own funds supporting the "financial security model, expand key technology R & D and demonstration application funding support channels, reduce the pressure of professional R & D funding.
Accelerate the development of standards and norms. The security market management department took the lead in integrating product R&D departments, engineering integrators, third-party testing organizations, and security professional users, and accelerated the research on the development of a standard system framework for security intelligent video analysis applications, from basic functions and overall technical performance requirements to application systems. The corresponding technical standards are compiled in terms of architecture, performance level classification, and detection and authentication methods. At the same time, through the establishment of an “Intelligent Video Analysis Product Certification Testing Center†and other entities, the authority and operability of intelligent video analysis performance authentication are improved. The market access and daily supervision system will be further improved to create a standardized and orderly market environment and promote “protection and protection†for the application of security intelligent video analytics.
With the promotion of safe city construction, the application of security video surveillance has become increasingly popular. With the rapid expansion of the security video surveillance system, traditional manual monitoring has been difficult to meet the actual needs of security in terms of abnormal goals, behavior, event recognition efficiency, accuracy, real-time and so on. The advanced concepts and innovative application modes of security intelligent video analytics application technology are refreshing. However, after nearly ten years of development, there is still a huge contrast between its actual application results and user expectations, and it has almost fallen into the deadlock of “a thousand functions of product features, poor depth of application, and chaotic market supervisionâ€. How to break through the bottleneck of the application of intelligent video analytics for security, requires market surveillance, product development, system integration, and professional users to work together.
I. Application bottlenecks and countermeasures of intelligent video analysis technology Brief introduction of security intelligent video analysis technology
Security intelligent video analysis technology originated in computer vision technology. It comprehensively applies techniques such as image enhancement processing. Based on the separation of goals and backgrounds, the target characteristics information is compared with pre-set templates or rules to automatically identify interest. The target, behavior, event, or data generates an alarm, and if necessary, the abnormal target can be automatically tracked and linked with other security facilities, significantly improving the overall efficiency and capability of security prevention and control. At the same time, the use of security intelligent video analysis technology in massive historical video information to achieve intelligent query search based on time, location and semantic characterization, can provide efficient verification means for the reconnaissance office.
Basic architecture
The overall architecture of the security video analytics application system mainly includes the following three modes:
Front-end embedded deployment mode: The intelligent analysis function module is placed in front of a video capture front-end device, and a shared camera image processing chip or dedicated high-performance DSP is used as an intelligent analysis engine to realize intelligent analysis of abnormal targets, behaviors, and events. This kind of architecture is not affected by the image transmission link, which eases the pressure of video transmission and background storage. At the same time, intelligent analysis using uncompressed original images has good real-time performance, strong pertinence, and ease of deployment on a large scale. The disadvantage is that it is difficult to support complex intelligent analysis applications, and due to the pressure of front-end processing, higher requirements are placed on the integrated heat-dissipation process and stability of the front-end device, and it is relatively easy to be constrained by environmental factors. In addition, the front-end embedded intelligent analysis software is difficult to upgrade, and to some extent, the difficulty and cost of installation, commissioning, operation and maintenance are increased.
Background server deployment mode: Intelligent analysis applications are implemented through dedicated servers (groups) or dedicated clients deployed in the security monitoring center. This architecture effectively solves the inconsistency of the front-end processing capabilities. It can implement complex target, behavior, and event intelligence analysis through powerful background processing capabilities; facilitates integration with other professional application systems and expands the scope of smart analysis applications; intelligent analysis engine background running environment Guaranteed, system work stability is greatly improved, and it is convenient for intelligent analysis software upgrade, effectively reducing the pressure of system operation and maintenance. In short, this architecture is suitable for phased fixed-scenario intelligent analysis applications (such as large-scale event security), but it requires high performance for back-end server configuration. The level of intelligent analysis is easily limited by the bandwidth of image transmission links, and real-time performance is not ideal. Large-scale smart analytics applications can't be done.
Distributed hybrid deployment mode: Place some relatively simple video pre-processing functions on the camera (or video server) to complete the extraction, packaging, and sending of target feature data. The background intelligent analysis engine is responsible for receiving target feature data, completing and presetting rules. Or modules analyze and compare and automatically generate alarms. This architecture reduces the pressure on the front-end intelligent analysis. The back-end server only needs to analyze the target data, and it is convenient to integrate with other professional application systems. It improves the efficiency and capability of intelligent analysis and is more suitable for large-scale, high-performance intelligent analysis applications. However, the system The overall cost of deployment and operational maintenance pressure cannot be underestimated.
basic skills
Abnormal target recognition tracking. Identify, lock and track abnormal dynamic targets that occur in a specific period of time and in a specific area, and automatically record their movement trajectories.
Abnormal events, behavior identification alarm. Identify abnormal events or behaviors such as cross-border, legacy, stranded, stranded, fall, trail, fight, steal, and gather within a specified area and automatically report an alarm. Feature data acquisition. License plate recognition, face recognition, traffic flow monitoring, and real-time monitoring of crowdedness in public areas.
Massive video smart search. According to the semantic description of the target, behavior, and event features, and the auxiliary information such as location and time of the comprehensive event, the video information needed in the massive historical video information is automatically searched to improve the efficiency of the investigation of the staff of the department's investigators.
Second, security intelligent video analysis technology application status
At present, the application of security intelligent video analysis technology mainly focuses on urban public security management (such as public security bayonet information identification, crossing vehicle inspection and control, Internet cafe face comparison, etc.), and large-scale event security (such as the Shanghai World Expo campus entrance and exit portrait collection and comparison, virtual Waterfront electronic fence, etc.), management of the prison and other business areas. The comparatively successful application model is based on feature data acquisition comparisons. At the same time, in the surveillance sites where monitoring scenes are relatively fixed and lighting conditions are better, the intelligence of the objects, events such as illegal cross-border, overtime detention, confrontation, abnormal aggregation, etc. is supervised. Analyzing the alarm application has achieved certain results.
As far as the overall application level is concerned, due to factors such as the effectiveness of intelligent analysis algorithms, the complexity of the background, and the uncertainty of lighting conditions, the intelligent video analysis application of abnormal targets, behaviors, and events in complex open environments is still more challenging than the practical requirements for distance security. Big gap. The license plate recognition rate of the security bayonet and city road crossing is relatively high (more than 90%), but there is still much room for improvement based on the depth application of license plate recognition. Static portrait matching technology can achieve more than 90% accuracy in comparisons when the quality of the comparison source and comparison target is relatively high (such as portrait photos provided by second-generation residents), and it has higher actual combat applications. Value, but the real-time comparison technology for face collection in the open environment is limited by many factors such as site lighting conditions, “blacklist†quality, and the degree of cooperation, and the accuracy ratio is about 60-70%. Currently, it can only be deployed as a large-scale event in real time. Aids.
Third, the main factors restricting the application of intelligent video analytics security
The lack of illumination at night in the open environment, bad weather (fog, snow, rain, dust, etc.), image compression processing, and the limited bandwidth of the network transmission link have caused image quality to decline, which has brought inherent difficulties to security intelligent video analysis; Complex anomalous behaviors and event modeling are difficult, and the corresponding intelligent analysis algorithms do not have high recognition performance. Targets are too close to the background or the background is cluttered, which leads to difficulties in target segmentation and feature information extraction. The moving target is blocked by the background and multiple moving targets obstruct each other. Target feature information is incomplete. The above factors are likely to cause adverse consequences such as false alarms, missed alarms, and tracking difficulties, which will severely restrict the actual combat performance improvement of the security intelligent video analysis application system.
The efficiency of the embedded intelligent video analysis algorithm based on D1 or HD resolution is not high, leading to the shortage of front-end DSP hardware resources, lack of front-end equipment integration, and lack of heat-dissipating technology to further limit the performance of DSP. The hardware processing capability and stability of the embedded intelligent analysis engine have been improved. It has become another major factor that restricts the application level of security intelligent video analytics.
Due to the fact that most of domestic security intelligent video analysis application products are from foreign manufacturers, and there is no scale application effect, the product pricing is generally abnormally high; manual calibration, background software debugging and other workloads during the implementation of the project are large, and some high-end product debugging even The need for direct participation of foreign manufacturers and technical personnel, coupled with the relatively high failure rate of embedded front-end equipment, makes it difficult to implement remote maintenance such as fault diagnosis and software upgrades, leading to high operating and maintenance costs for application systems. The above factors result in the low cost performance of security intelligent video analysis applications, which severely limits the effective promotion of security intelligent video analysis applications.
Mechanism level
At present, the construction of the security video surveillance system is still at an extensive stage of scale expansion. As the main unit of construction, security users are still paying more attention to the "monitoring coverage rate," and the degree of attention to prior warning is not enough. The direction of intelligent video analysis is The investment in R&D funding is also obviously insufficient.
Security intelligent video analysis product development lacks effective mechanisms. On the one hand, most security users do not seriously sort out application requirements. On the other hand, the product R&D department lacks an in-depth understanding of the security industry, resulting in intelligent video analysis products that have a thousand functions and lack of targeting. Sex. At the same time, most system integrators do not give full play to the bridge between product R&D and users. Some integrators even exaggerate the functions of intelligent video analysis products and seriously damage the market image of intelligent video analysis products.
In addition, from the perspective of market supervision, due to the lack of authoritative standards certification system, users can not effectively measure and control the product performance, and seriously affect the user's confidence in the application of security intelligent video analysis.
Fourth, the main countermeasures to promote security intelligent video analysis applications
Focus on the support of local R&D forces and concentrate on key technologies.
In order to cope with the current situation in which security and intelligent video analytics applications rely heavily on foreign technology, and at the same time effectively stop the inefficient research and development of simple imitation and repeated investment, it is necessary to carefully organize local professional research institutes to focus on basic and common key technologies to break through security. Intelligent video analysis depth application creates conditions.
Efficient intelligent video analysis algorithm: Improved target segmentation, feature extraction, target recognition, and dynamic tracking algorithms based on complex scenes (such as varying lighting conditions, closeness of target and background colors, background clutter, targets blocked by background, and changes in monitored field of view) Performance improves the capacity and effectiveness of static object libraries and dynamic rule bases, reduces the occupation of hardware computing resources, and improves the accuracy and real-time performance of automated analysis of complex targets, behaviors, and events for open, complex, and large-scale monitoring areas. At the same time, the overall structure of the intelligent analysis software is increased to effectively take into account the real-time and robustness of the intelligent video analysis software, and meet the intelligent analysis requirements for different resolution video streams such as SDTV and HD, making the intelligent analysis software architecture more rational and efficient; Based on the automatic calibration technology of target characteristics, the ability to adaptive "learning" of back images is enhanced, the probability of false alarms is reduced, and complicated manual calibration is avoided, the difficulty of system debugging is reduced, and the maintainability of the system is improved.
Image acquisition and comparison technology: Further improve the algorithm for image acquisition and comparison based on open space and no-match conditions, eliminate the influence of unfavorable conditions such as complex and changeable on-site lighting and free movement of personnel, improve the accuracy of real-time comparison of the system, and improve In fact, the application value of warfare is used to improve the availability of backstage portrait storage data by automatically eliminating algorithms such as redundant portrait data; and to improve the reliability and real-time performance of automatic search and query comparison of static heterosexual portrait targets based on large-capacity portraits. In addition, we will increase the research and development of portable and transportable portrait collections compared to terminals, increase suspicious target maneuver control capabilities based on portrait collection and comparison technologies, and vigorously expand the application of portrait acquisition and comparison technologies in the practical application.
Massive video intelligent search technology: further improve the performance of the video semantic analysis (including the semantic feature extraction of video, multi-granular event representation, etc.) algorithm performance, and improve massive video intelligent search based on target or event semantic description and event time, place, region, etc. Real-time and accuracy, while providing mobile target trajectory analysis and display; accelerate the development of portable intelligent video search and comparison of terminal equipment, facilitate the search and investigation of departmental reconnaissance personnel on the independent storage of historical video search, improve the efficiency of video forensics.
Blurring image processing technology: Based on the selection of low-lightness, high-definition cameras and bandwidth of the transmission network as much as possible, further increase the development of fuzzy image processing technologies based on image enhancement, image restoration, and super-resolution reconstruction to improve the weather. The blurry image processing capabilities caused by the target motion, the defocusing of the optical lens, etc. lay the foundation for the subsequent application of intelligent video analysis. Increase demonstration application efforts to improve the level of intelligent video analytics applications in actual combat.
The key to the breakthrough of security intelligent video analytics level lies in application. Only through actual combat can system performance be truly verified and continuously improved. As a security department, we can neither regard intelligent video analysis technology as an omnipotent "magic bullet", and we should not adopt a passive and passive attitude. Instead, we should actively and steadily improve security based on the application strategy of "demand traction and demonstration first." Video analytics application level. According to the development status of the security intelligent video analysis technology, combined with the urgency of security business requirements, priority can be given to the following demonstration applications:
First consider the occasions where the monitoring scene is relatively fixed, the lighting conditions are easily secured, and the target and behavioral information are highly identifiable (such as inside large-scale event security areas, indoor ATM machines, and prisons, detention centers, detention centers, interrogation rooms, etc.) Carry out automatic analysis of alarming applications of abnormal targets, behaviors and events to improve the safety management and control capabilities of closed environmental areas; develops and implements massive data back-end mining and intelligent association analysis applications such as public security bayonet information identification, electronic*, and city-level crossings, vehicle and vehicle inspection and control, etc. Security check and control capabilities of security bayonet and city road crossings; promotion of large-scale event security zone entrances and exits, railway station high-speed trains (EMUs) passenger self-entry passages, metro station underground pedestrian passages, and other area portrait collection and comparison demonstration applications. The personnel in the public agglomeration area have suspicious target deployment capabilities.
Secondly, intelligent target analysis based on complex backgrounds or scenarios, behavior, and event intelligence analysis are the difficulties in the application of security intelligent video analytics, and it is also a key area for security application. Therefore, under the premise of effective breakthroughs in the development of key technologies such as efficient intelligent video analysis algorithms, it is necessary to selectively select intelligent video analysis depth applications in areas with high crowd concentration and complex public security situations in order to enable intelligent video analytics applications. We will make substantial progress as early as possible in the critical areas of security operations and create necessary conditions for large-scale application of security intelligent video analytics.
V. Actively explore and practice, strive to break through the traditional mechanism constraints
Innovative product development mechanism. The performance of intelligent video analysis products is highly related to the characteristics of industrial applications. Considering the particularity and sensitivity of the security industry, if the traditional market independent R&D model is used, it is difficult for professional R&D institutions to accurately understand and grasp the complex business needs of security departments. It is also difficult to truly meet the security requirements. In view of this, it is necessary to innovate the “triangulation, integration, and user†tripartite cooperation mechanism in the form of a “joint key laboratory for smart video analytics application technology†based on the principles of “benefit sharing and risk sharingâ€. In the cooperative R&D process, security users mainly actively involved in the product R&D process by combing application requirements, organization and demonstration construction, and coordinated performance improvement; professional R&D institutions are working hard to improve intelligent video analysis algorithms based on a thorough understanding of security business requirements. Performance improves real-time performance and reliability while reducing the probability of false alarms and false negatives in the system. At the same time, it improves the cost-effectiveness of intelligent video analysis products. System integrators maximize their performance through camera selection, supplemental lighting, monitoring angle adjustment, and calibration. Provide high-quality surveillance images, highlight monitoring target feature information, and create basic conditions for intelligent video analytics applications. Through the above-mentioned cooperation mechanism, it promotes three-way benign interaction and leads the right direction of security video analytics R&D and application. More importantly, through the sharing of intellectual property rights and research results by security users, the cost of system construction can be effectively reduced, and the inherent enthusiasm, initiative, and creativity of security users can be fully mobilized to accelerate the promotion and application of security video analytics applications.
Expand funding support channels. At present, smart city construction based on core supporting technologies such as the Internet of Things has entered a substantial start-up phase in many domestic cities. Smart security is one of the key applications for the demonstration of Internet of Things, and it has also been included in the scope of smart city construction. Therefore, in order to effectively solve the problem of high price for security intelligent video analysis products due to insufficient application scale, it is possible to build “government scientific research funding support through security users, R&D institutions, and integrators jointly applying for key technology research and demonstration application projects. The company's own funds supporting the "financial security model, expand key technology R & D and demonstration application funding support channels, reduce the pressure of professional R & D funding.
Accelerate the development of standards and norms. The security market management department took the lead in integrating product R&D departments, engineering integrators, third-party testing organizations, and security professional users, and accelerated the research on the development of a standard system framework for security intelligent video analysis applications, from basic functions and overall technical performance requirements to application systems. The corresponding technical standards are compiled in terms of architecture, performance level classification, and detection and authentication methods. At the same time, through the establishment of an “Intelligent Video Analysis Product Certification Testing Center†and other entities, the authority and operability of intelligent video analysis performance authentication are improved. The market access and daily supervision system will be further improved to create a standardized and orderly market environment and promote “protection and protection†for the application of security intelligent video analytics.
Other Painting Tools,Painting Tools Online ,Painting Tools Art,Painting Tools List
Paint Brush, Roller Brush Co., Ltd. , http://www.nbpaintbrush.com