Video Retrieval - Analysis and Management of New Security Data
Video retrieval is to find the required video clips from a large amount of video data. With the development of computer technology and network technology, video retrieval technology has been widely used in video on demand, digital TV, digital library distance education and telemedicine. And other fields.
In the field of security, the construction of safe cities, smart cities, and the construction of information technology in the field of public security have generated a large amount of surveillance video data. At the same time, the development trend of high-definition and intelligent security industry, the number of video data is being indexed. Level of growth. However, video data is different from other traditional information data in the security field. In the past, traditional information, after years of development of database technology, has formed a complete management platform based on information management system, which can quickly and accurately search and query the data of interest, such as population information management system, alarm platform, Vehicle management system, etc. However, there is no such effective management system for video data, and it is not convenient and quick to query and use the data that exists. The development of video retrieval technology is to solve such problems.
In the field of security video search technology is mainly used in image detection, the role of video data in the detection of cases, also prompted the image detection to become a large new police. The specific applications of video retrieval mainly include video enrichment and feature retrieval, while feature retrieval usually requires the support of intelligent analysis and image enhancement technology.
The information density contained in the surveillance video data is very low. For a specific requirement, the useful data in the 24-hour video data may be only a few seconds. However, the information value of the data in these seconds is very high. Video enrichment technology is used to effectively increase the information density of video data. It separates the moving target of the video from the background, and then overlays the background in time and position to form a new video, which can maximize the video. Preserve environmental information while compressing useless information in time and space. The feature retrieval technology refers to indexing the features of the target in the video content, and then searching for the video by describing the feature or the features displayed by the sample image, finding a video segment that matches the feature, and quickly obtaining more information from the device to improve efficiency. , saving labor costs.
Application and status quo - initial results, urgent need to improve
Many security enterprises or related institutions in China have successively launched their own video retrieval systems. The specific work processes and principles are similar. The first step is to collect and centrally store the video data that needs to be retrieved for quick recall by the video retrieval system. Some manufacturers have developed specialized devices to quickly copy video data from the hard disk to improve the acquisition speed. The second step is to formally perform video retrieval. One is to concentrate the data on the summary, and the concentrated video data is provided to the investigator. Manual troubleshooting can also be used for feature retrieval; the second is to retrieve video data through user-provided features.
By looking at the condensed video, the case handler can greatly save the time for troubleshooting the video. When a suspicious target is found in the condensed video, you can quickly open the corresponding original video clip by clicking on the target on the video for detailed viewing. The case handler can further initiate the search by providing target features such as person or car, target color, height, direction of movement, speed of movement, pedestrian dressing, gait, or by providing a sample.
The video retrieval system first decodes the undecoded video data through an efficient decoder, and then extracts and extracts the moving targets through the background modeling algorithm, and then extracts and analyzes the focused features of the moving targets through the intelligent analysis algorithm. The analysis results are properly described, stored in the database, and compared with the features of the requested search. Finally, the high-correlation target is displayed as a result by snapshot or other means, and the original video clip can be quickly located.
Some manufacturers also analyze the basic characteristics of some moving targets, such as target categories, colors, speeds, etc., in real time during video recording, and index the video content through these features and store them in their own devices. In this way, when the feature retrieval is needed, the related video clips can be located only by comparing the features provided by the user with the index information of the video content, without the need to process and analyze the video again, which can effectively save analysis time. However, the standards for indexing and description of video content of different vendors are different at present, which will result in the problem of system docking and video content index sharing between different vendors, and the role of the system cannot be fully utilized. MPEG-7 standardizes the description of audio and video features, gives a direction for solving the problem, and further standardizes and standardizes the characteristics of security video data.
At present, the video retrieval system being applied in China can only index and retrieve some basic features of the target. The current video retrieval system cannot complete the task of "finding a girl with iPhone6" because of the resolution, With the influence of multiple factors such as clarity and video angle, we can't tell the model of the phone. We don't even know if there is anything in the target. We can't even judge whether the goal is a boy or a girl because of the fast changing era of this popular factor. . Some of the amazing intelligent analysis effects shown in some film and television works can only provide researchers with some research ideas and directions.
In addition to being able to search for some of the basic characteristics mentioned above, what we can do now is to retrieve the license plate information, which is due to the maturity of the license plate recognition technology and the construction of the intelligent transportation system. For tasks like face recognition, for general surveillance video, even high-definition bayonet is difficult to achieve high accuracy, or combined with manual comparison to really play a role, the human eye can accurately distinguish and identify The goal of video retrieval system is not necessarily able to achieve, to achieve the ability of the human eye to distinguish, that is, after all, is the highest realm and ultimate goal in the field of machine vision, there is still a long way to go. Seeing the hidden propaganda of some products, the suspects in the video changed clothes, wore hats, changed vehicles, and the retrieval system could accurately retrieve the target. The author was ignorant and skeptical.
We introduce a real case, which is the most common application mode of video retrieval technology in China's security field. In April 2013, a major masked robbery occurred in a certain place. After analysis by the task force, it was considered to be a carefully prepared and premeditated case. The suspect may have investigated the crime scene several times in advance. In order to solve the case as soon as possible, the local public security bureau's map investigation brigade transferred all kinds of surveillance videos around the crime scene, including traffic surveillance video, public security surveillance video and own surveillance video of nearby stores. The total duration is more than 2000 hours, hoping to get from it. Identify suspicious targets and provide clues to solve the case.
Traditionally, it takes several days for several police officers to view the video at the same time. Through video enrichment technology, a police officer can check out suspicious persons or vehicles in the video within a few hours, and the same criminal investigator completes the investigation. It can also be more sensitive to the recurrence rate of suspicious persons in the picture, which saves the inspection time and reduces the probability of omission. When the suspicious vehicle is filtered out, the license plate identification information is used to form the trajectory of the vehicle in the system, and the reference scheme is provided for reference. Through the action law of the vehicle, the area where the vehicle stays for a long time is located, which may be the temporary residence of the suspect. , then call the video around the area for summary investigation, combined with the suspicious target comparison of the crime scene, and finally locked the suspect.
The era of big data - the opportunities and challenges of video retrieval technology
The scale of security infrastructure construction is constantly expanding. The scale of real-time data has reached the order of PB. At the same time, video data from the consumer field is also providing data support for security. Everyone's mobile phone may upload images to the public security platform. Information, providing case clues, to quickly and efficiently filter effective information from these massive information, video retrieval technology needs to be improved in two directions: first, it needs to improve the processing power and efficiency of video data, to meet the level of massive data. Process data more accurately and quickly, extract and describe the content of the information, and efficiently index and store to meet subsequent data information retrieval, analysis and mining services. Secondly, it is necessary to improve the current data processing technology and storage architecture from the perspective of architecture, and introduce new technologies for big data services to meet the retrieval requirements of security data information in big data environment.
At present, the distribution of video data in the security field is mainly distributed and stored independently on the servers of various organizations. The application of video retrieval is mainly performed by an independent server, and as the amount of data rapidly expands, this This approach will certainly encounter bottlenecks in data movement and computing power in the near future. Big data-related cloud computing and cloud storage technologies are applied to security video retrieval services, which can solve the problem of data movement, and fully exploit the advantages of distributed computing to discover sufficient computing power.
In the security field, big data and video retrieval technologies are actually promoting each other. Video retrieval technology provides technical support for the effective application of big data. Big data also forces the efficiency, accuracy and coverage of video retrieval technology to improve. These two technologies will gradually improve and unify in the mutual integration.
Shenzhen ChengRong Technology Co.,Ltd. , https://www.nblaptopstandsupplier.com