Application of Video Data Mining Technology in Monitoring System

In recent years, with the widespread installation of video surveillance systems, the capacity of video data captured and stored by surveillance systems is increasing at an alarming rate. From an idealistic point of view, these videos contain a large amount of information in the real world, but from a practical point of view, it is very difficult or even impossible to rely on manual processing of tens of thousands of video data sets to obtain information from them. As a result, the vast majority of the video data captured and stored by the surveillance system becomes data that is not stored on the hard disk, leaving us in a dilemma of data explosion but lack of information.

The reason why video data cannot be used directly by us is because there is a "semantic gap" between the two, which is the difference between low-level image features understood by computers and high-level semantic information understood by humans. For example, when watching a surveillance video, humans can quickly combine prior knowledge to determine more detailed information such as the pedestrians, the people talking, and even the relationships between pedestrians and emotions in the video. The computer can only obtain information. Image patch, area texture, or motion direction image characteristics.

Data Mining Technology Sets up Man-machine Communication Bridges

Data mining technology is a bridge that lies above the "semantic gap", enabling us to obtain semantic information that can be applied from image feature information that cannot be directly understood. However, at present, the mining of video data is still a difficult problem in the data mining technology field. Unlike text data, video data is an irregular data format with a very large amount of information. It does not have rules such as grammar and paragraphs of text data; The amount of information contained in the video is quite large, and it has high difficulty for feature fusion and information extraction of the video data. So overall, video data mining technology is still in its infancy, but some technologies have reached a stage of more mature scale application, such as license plate recognition technology, video intrusion detection technology, and so on.

The user needs of video data mining solutions How to extract from the massive video data to the information we can apply, or even summed up the knowledge, is undoubtedly the monitoring system users in various industries are urgently needed to solve the problem. However, video contains a very large amount of information. Customers in different industries have very different ways to extract and use video information. This requires suppliers of monitoring technology to provide different video data mining solutions for customers in different industries. For example, users of the public security industry need to obtain timely information on security anomalies in daily security management, such as fighting incidents or crowded incidents, so that they can deal with them in a timely manner; they must have a lot of video when conducting criminal investigations. To perform target search, you need to obtain the target's identity information from the video, such as personnel identification information and vehicle license plate information. The users of the expressway industry need to obtain vehicle license information at the toll collection office and obtain abnormal event information such as congestion incidents and parking in the road surveillance video. Statistical information such as traffic flow and average speed must also be extracted to achieve management optimization. There are also some video information needs of users in various industries, such as video quality information, that is, the amount of information that the current video equipment is operating from the video data is normal, which has important practical value for the operation and maintenance of various industry monitoring systems. .

According to the actual application requirements and different application methods, the information mined in video can be divided into five categories: event semantic information, target identity information, target image feature information, video statistics, and video quality information. The event semantic information refers to the event information described in the available language from the video. For example, someone breaks into the area, someone runs, and a group gathering event occurs. This type of information is presented to the user in real time in the form of an alarm. The user can use this to The class information judges and processes abnormal events in real time. The target identity information mainly refers to the personnel identity and the vehicle license plate information. The user uses such information in the manner of an alarm or retrieval, such as a vehicle blacklist alarm or a suspect photo retrieval. The target image feature information refers to the target image features that can be described, such as red cars, people wearing black and white striped clothes, etc. The user can use this type of information in criminal investigation work to quickly locate the target in the massive video data. The video statistical information refers to the long-term statistical data obtained from the video, such as the traffic volume of the mall and the traffic volume of the traffic arteries. The user can use this type of information to optimize the management work. The video quality information refers to the information that describes the video quality anomaly obtained by diagnosing the video quality. For example, the video is blocked, the video is out of focus, and the video is color cast. The user can use this type of information to perform the operation and maintenance of the monitoring system.