Analysis and Application of Monitor View Extraction Technology

The rapid development of video surveillance technology in intelligent analysis benefits from the popularization and application of smart chips, algorithms, and technology architecture in business applications. With the diversification of the business caused by the technological differences in competition, but also makes security companies in the development of the entire security monitoring technology has its advantages.

Industry Background and Demand Overview

Thousands of surveillance cameras in public places in cities are constantly monitored and recorded around the clock, resulting in massive amounts of video data. It is very difficult and time consuming to manually search targets for these massive videos. Some monitoring videos of key venues have a strong real-time and efficient demand for target search. At present, the use of artificial means is inefficient. Even if video-enrichment abstracts and other technologies are used to process historical videos and images, and the procedures are cumbersome, sudden emergencies will often delay the timing of investigations, leading to relevant intelligence investigations and cases. The response speed of detection is not enough to meet the requirements of prior high-risk early warning and immediate event handling. In the video surveillance collaboration* detection case, the most ideal situation is that once there are important events, you can quickly find the key target people, vehicles, objects and other view clue information, and then carry out human and vehicle deployment can also be real-time Early warning.

Based on the above background and requirements, the view extraction technology can meet these requirements. The goal is to use real-time, high-efficiency, automatic, and intelligent methods for the traditional video cues based on human tactics to search for or process historical video images afterwards. Technical realization.

View Extraction Technology Analysis

Surveillance cameras produce a vast amount of raw, rough information that is often used to store or accept surveillance of full-time personnel. If the video information is handled manually, it will become monotonous and boring. At the same time, most of the time, the original video does not include the target events that the case concerned. How the camera is used for effective scene monitoring and real-time pre-warning as required is the research objective of the video surveillance problem. This requires a video surveillance technology that automatically captures the description of what is happening in the monitoring scene and takes appropriate actions based on video analysis.

Although different video surveillance systems have different monitoring requirements, they all need to deal with some common problems during the monitoring process. We begin by analyzing the working methods of monitoring personnel to discuss these common issues. In the face of video surveillance images, monitoring personnel usually only pay attention to certain specific objects in the scene. For example, only focusing on the new target (strange person) in the monitoring scene will not pay attention to the scene change of the monitoring scene or the change of the position of the object. The object (strange person) is separated from each frame of the video image, and the same object is tracked for a long period of time. This is the problem of object detection, classification, and tracking. The monitoring staff analyzes and analyzes a series of information generated during the tracking of the target, and then assigns a description of the behavior to the movement of the target, and determines whether or not the event occurred in the monitoring scene according to the behavior description. Give an exception report. This process for the video surveillance system, is based on the target object is identified and confirmed, based on the analysis of the target's behavior, according to the set conditions to classify the system's understanding of the behavior, give a description of the event behavior in the monitoring scene. In summary, through the analysis of the actual work flow and working methods of the monitoring personnel, we know that the major issues involved in the video surveillance process include: identification of target objects, tracking of target objects, and behavior analysis of target objects. The core of the problem is image classification and image search. Image classification tasks such as: human and vehicle goals, vehicle brands, vehicle color recognition, vehicle type identification, etc.

The way of view extraction technology

The view extraction technology realizes the real-time structured extraction of massive views, separates people, vehicles, and objects, extracts effective structured data, and performs in-depth structured processing of cars and faces in the view, real-time perception of the city's High-risk personnel, high-risk vehicles. Structured high-value information can be stored for a long period of time, making post-discovery search targets more convenient and saving video surveillance time and labor costs.

Human separation

Separation of video and car objects Structured processing of video/video of platforms/equipment, and the movement targets in video are divided into people (pedestrians, bicycles, battery cars or motorcycles, tricycles), cars, and objects, and basic targets are extracted. Information, such as subject color, target size, target speed, direction of movement, time, etc. According to these extracted target information, target search can be freely combined and the display result is displayed in the form of a view, a list or a map. Each record corresponds to the original video clip for detailed viewing. Structured important clues and high-value data can be visualized into the library after analysis by the civilian police. At the same time, the extraction system can also obtain high-value information from other business systems after viewing the library for the purpose of concatenating the same person and vehicle clues. Show on the map.

Vehicle/face recognition

The vehicle and face recognition is based on the vehicle resources separated from the existing bayonet electric police equipment, bayonet platform, and human and vehicle objects, realizing the secondary extraction and analysis of the intelligent features of the passing pictures, and performing structural analysis and extraction. (Including: Vehicle brand, model, year, color, category, abnormal behavior, etc.), as well as real-time search and analysis of vehicle passing information, such as the unique characteristics of the vehicle (such as annual inspection marks, tissue boxes, sun visors, ornaments, and pendants). Mainly include: license plate recognition technology, vehicle model identification technology.

License plate recognition includes quickly locating the license plate position and accurately identifying the technologies required for a series of intelligent traffic fields such as license plate characters, license plate colors, body colors, front logos, and vehicle types.

The vehicle model identification technology currently uses international advanced deep learning technology combined with vehicle size, body color, and license plate position and color, etc. It can accurately identify and identify hundreds of car brands and support thousands of specific models of vehicles. Accurately identify, and the algorithm does not depend on the * database, has a flexible application scenario, to provide more refined information for * criminal investigation work. (Supports analysis output of 200 brands, 3000 subdivisions, and local features.)

Vehicle features display

Difference from condensed abstract

The abstraction of the face of demand is based on real-time structuring, so the application focuses on the real-time structured requirements of key areas, and the summary of the summary is abstracted and analyzed for all subsequent videos.

Extraction For the video structuring, in addition to the human-vehicle separation function, there are vehicles that perform secondary analysis and feature dispatching, while the enrichment summary focuses on the separation of people and vehicles, which cannot be achieved for car face recognition and feature deployment.

Video extraction technology application

The products developed using the view extraction technology are mainly located in the business applications of *. The applications can be divided into the following types:

Focus on real-time structuring in advance

Supports configuration through electronic maps, real-time structuring of points in accordance with intelligent analysis around key locations, support in accordance with full time periods, important time periods (eg 7-9 am school, 11-13 noon bank, afternoon 3-5 schools) etc. to configure. Once an incident has occurred, the suspect and vehicle information can be quickly located by searching structured information.

Instant case processing

For scenes that need to be tracked by video immediately after receiving the alarm, support the direct extraction of the points around the location via the electronic map box, and can control the suspects and vehicle features (red clothes - personnel, black sedan) Such as) the way to quickly grasp the trajectory of suspects.

At the same time, it is also possible to gradually adjust the camera for structural analysis based on the direction of movement of suspects and vehicles to track suspect trajectories in real time.

Afterwards, trajectory reduction

The time-space analysis can be performed on the map, and the personnel and vehicle information extracted by the system analysis can form the space-time trajectory of the target, and help the police in handling the case to analyze and judge.

Industry Technology Application Status

At present, mainstream security product solution providers mainly adopt the intelligent analysis technology of video information. Commonly used intelligent video analysis technologies achieve target detection, recognition, and feature refinement through algorithms such as target extraction, target tracking, and feature extraction. Features. Security companies are also launching products based on these technologies for business applications, which fully demonstrates that this technology can bring convenience to business applications.