Video surveillance application based on multi-target recognition and tracking technology

Implementation of multi-target recognition and tracking technology

This article refers to the address: http://

Single ball machine tracking technology

The realization of multi-target recognition and tracking technology of HC Security Network is based on the single-ball machine intelligent tracking technology system. Therefore, manufacturers with multi-target recognition and tracking products on the market often have corresponding intelligent trackballs. Machine product series. The key points of the intelligent tracking technology of the single ball machine are mainly the intelligent algorithm and the ball machine control technology.

The key of the intelligent algorithm is to analyze and process the image sequence containing the moving target. The core technology is the detection, tracking and recognition of the moving target in the scene, as well as further behavior analysis and event detection. The purpose of target tracking is to correlate the same moving objects in different frames in the image sequence by processing and analyzing the video data, so as to calculate the motion parameters of the target, such as position, velocity, acceleration and motion trajectory.

Video surveillance application based on multi-target recognition and tracking technology / Jiang Hao's existing motion target detection methods can be summarized into four types: background subtraction method, time difference method, optical flow method and motion energy method. These algorithms have been widely used in various types of intelligent video analysis products, and their performances include multi-function detection functions such as smashing, regional intrusion, electronic fence, flaw detection, item retention, and item moving.

To achieve tracking of the target, some mathematical tools need to be applied to create corresponding matching relationships based on position, velocity, shape, texture, color and other related features between successive image frames. At present, in terms of tracking personal objects, there are tracking body parts such as hands, faces, heads, legs, etc., as well as tracking the entire human body; in terms of tracking angles, there is a single angle of view corresponding to a single camera, multiple angles of view corresponding to multiple cameras, and all Azimuth perspective; it can also be classified by tracking space (two-dimensional or three-dimensional), tracking target categories (people, cars, etc.), camera status (motion or fixed).

The ball machine control technology mainly refers to the control of mechanical moving parts such as the angular velocity of the pan/tilt, the magnification and speed of the lens zoom, and needs to be combined with the intelligent tracking algorithm to achieve smooth tracking, appropriate magnification, and appropriate target ratio. Visual effects.

Multi-target tracking technology

Multi-target recognition and tracking technology is based on the intelligent tracking of single-ball machine. It can simultaneously realize intelligent identification and tracking of multiple active targets in a wide range, and intelligently track individual targets.

Multi-target recognition and tracking technology In the application, a fixed camera is usually used to analyze the intelligent behavior of targets in a wide-area range, and multiple targets monitored at the same time are sorted according to the established strategy, and commanded in order. The intelligent trackball machine tracks the monitoring targets one by one. Compared with single-target tracking, the key point of multi-target tracking technology is data association problem, which is to establish a unified coordinate system, so that the fixed camera can transmit the coordinate information of the target to the tracking dome to achieve linkage tracking.

The process of multi-target tracking can be divided into the following parts:

Data association: establish a correspondence between observation data and target

Common methods include nearest neighbor algorithm, joint probability data correlation filter, and multiple hypothesis tracking algorithm.

State estimation: each target is estimated based on its corresponding observations

Product form of multi-target recognition and tracking technology

Bayesian theory-based methods are usually used to transform multi-target tracking problems into tracking processes for multiple single targets and to establish corresponding state space models. Assign a single-target tracker to each target, track each target independently of each other, and design some special methods to handle interaction and occlusion issues between targets.

Coordinate transfer: establish a unified coordinate system between the main camera and the dome

In the multi-target monitoring scene, the position and motion trajectory information of the target are extracted, sent to the slave camera, and the camera locks the target according to the position of the target and the motion trajectory information.

In addition, in the practical application, multi-target tracking technology also needs to focus on optimizing and improving the following aspects: improving the efficiency of the algorithm to achieve tracking of as many targets as possible; improving the anti-jamming performance of the algorithm to mitigate light changes and shadows Common interfering factors such as occlusion between targets; need to be able to prioritize each target so that the dome can switch to a target with a higher threat level in time when tracking.

Product form of multi-target recognition and tracking technology

Multi-target recognition and tracking products are currently not available on the market, and are mainly divided into two types: gun-ball machine integrated and stand-alone products (including NVS, DVR, smart box, etc.).

The integrated gun-gun product refers to a multi-target tracking system that can be completely installed and used independently by embedding a wide-angle camera above the dome through a unique mold design. Since the position of the gun is fixed in the dome, it is easier to implement in the algorithm, and has good precision in scene matching and target coordinate transmission.

Stand-alone products are the most common form of products on the market today. This function is usually provided in NVS or DVR, or a dedicated video processing analyzer is designed. The gun and the ball machine are connected through the field wiring, and the master-slave multi-target tracking function is realized through the configuration of the platform software. The advantage of the stand-alone product is that it is flexible in application and can be compatible with all kinds of cameras from different manufacturers, including SD, HD, analog, network type guns and ball machines. It is not only suitable for new monitoring systems, but also for existing monitoring. The upgrade of the system has a high degree of flexibility in the design of the system and the construction of the system, and may be further developed in the future.

Development and improvement of multi-target recognition and tracking technology

As an emerging technology, multi-target recognition and tracking has been continuously improved and improved according to the needs of the market. We can try to analyze the main development direction of the technology in the future from the perspective of big security.

Smarter target prioritization

The ability to prioritize the target determines whether the current screen can always track the target with the highest risk level. When multiple targets are avoided, the dome will only follow the simple rotation sequence, thus effectively improving the application value of the product. An effective way to improve this ability is to set up multiple warning lines and regional identification rules in the scene. Second, you can consider the analysis techniques of introducing target behavior characteristics, and take the target characteristics such as bag, certain clothing color, and rapid running. The basis for determining the priority.

More effective against various disturbances in the environment

Multi-target monitoring systems are more likely to be used in outdoor open areas such as airports, plazas, and roads. They often face poor monitoring environments such as local glare, night, wind, heavy rain, haze, and ground water. Intelligent algorithms are needed. Continuously make targeted adjustments and improvements to ensure effective tracking in all real-world environments.

In addition, multi-target mutual occlusion, multi-target motion state differences and changes, is also a major factor affecting tracking performance. Whether it is possible to introduce binocular recognition technology and overcome multiple occlusions by multi-angle tracking, this is also a possible solution.

Single-gun machine + multi-ball machine, ball machine + ball machine combination application

The combination of a single fixed camera and a single billiard machine is currently the standard application.

In the large-scale monitoring system, for a large number of targets such as airports and plazas, and monitoring a wide range of scenes, if a combination of single-gun and multi-billiard machines can be realized, it will effectively improve multiple high-security targets in complex environments. At the same time monitoring capabilities, this will greatly improve the technical level of the current product.

In addition, the linkage tracking and relay tracking between the ball machines is also a hot spot technology in the industry. With the use of electronic maps, it is easy to achieve seamless tracking of scenes such as parks, and achieve high-level security requirements such as target trajectory depiction and crime behavior warning.

Conclusion

Multi-target recognition and tracking technology is also an emerging technology type in the surveillance market, but it has already shown good market development momentum and broad market prospects. It is believed that with the continuous deepening of technology research, more and more manufacturers and products will join this field to jointly promote the vigorous development of this market.

5050 Single Color Led Strip

5050 Single Color Led Strip ,Single Color Led Strip,Single Color Led Strip Lights,Single Colour Led Strip

NINGBO SENTU ART AND CRAFT CO.,LTD. , https://www.lightworld-sentu.com

Posted on