Professional image technology and
machine vision development and manufacturing

24-hour service hotline:
Current Lcation:Home / 产品方案 / 机器视觉知识
Machine vision subverts traditional technologies to help video surveillance systems become intelligent
The author:SIPO  Time:2018-04-02  Read: 152

Vision is the main source of human perception of the objective world. After the theory of signal processing and the emergence of computers, people tried to use cameras to obtain environmental images and convert them into digital signals. The whole process of visual information processing was implemented using computers. This kind of research goal that enables computers to recognize the surrounding environment information through one or more images is machine vision.


As a representative application of machine vision, video surveillance systems based on cameras have been widely applied in various fields such as security, transportation, buildings, and industries. To understand what changes machine vision brings to video surveillance, we can analyze it further from several major areas of machine vision.

Target Recognition

    Target recognition technology and stable tracking method are one of the key factors in the development of machine vision. It has been widely used in many fields, such as fingerprint identification, face recognition, iris recognition for identity verification, and license plate recognition in applications such as intelligent traffic management, vehicle detection, and parking lot management.

    A target recognition system should have the ability to detect, classify, and identify targets in complex backgrounds and various weather conditions so that the target can be continuously tracked.

    In recent years, the target recognition technology has gradually moved from theory exploration and laboratory simulation to practical application. Its technology method has also evolved from the classical statistical pattern recognition to the identification method based on knowledge, model, multi-sensor information fusion and artificial neural network.

    Target tracking

    Motion target tracking is the process of determining the position of the same object in different frames of an image sequence. Its main working method is to select good target features and adopt appropriate search methods. According to the matching principle, existing tracking methods are divided into tracking based on models, regions, features, and active contours.

    Model-based tracking

    Model-based tracking builds a model of the tracked target with certain prior knowledge, then matches the tracking template and updates the model in real time. There are three types of traditional moving object expression methods:

    1. Line chart method: The essence of the target movement is the movement of the main frame, so this expression method approximates all parts of the object with a straight line.

    2. Two-dimensional contour: The use of this expression method is related to the projection of the object in the image.

    3. Three-dimensional model: The three-dimensional model of the generalized elliptical cylinder, ball, etc. is used to describe the structural details of the object. This method often needs to match the three-dimensional model between the contacted image frames to obtain a quantitative description of the motion of the object. Therefore, more parameters need to be calculated, and the calculation process of the matching process is larger.

Feature-based tracking

    The pre-extracted motion area is used as a matching target template to set a matching metric, then the search target image is matched in the next frame of the image, and the metric taken value is determined as the best matching point. This method is an opportunity area. track.

    Due to the extraction of a more complete target template, this method can obtain more abundant image information than other tracking algorithms, and is therefore widely applied to tracking a smaller target or a target with poor contrast.

    Activity-based tracking

    A closed parametric curve is used to express the contour of the moving target. In the characteristic field constructed by the image, dynamic iteration is performed by minimizing the energy with the curve function as a parameter, so that the contour can be continuously updated automatically. Compared with the area tracking method, this method has a small amount of calculation. If each of the moving objects can be reasonably separated and the contour initialization is started, even if there is partial occlusion, continuous tracking can be performed.

    Visual analysis

    Visual analysis technology is to further acquire the target’s appearance time, movement trajectory, color, and other information through target recognition and target tracking. Through the analysis of the above information of each target, the dangers, violations, or suspiciousness found in the video are found. Goals, and real-time alarms, early warnings, storage, and retrieval of these behaviors and targets.

    In the application field of visual analysis, the most important ones are intelligent video surveillance and intelligent video retrieval technology. The application technology of the two is similar, the main difference is that: intelligent video surveillance is real-time processing of the collected video at the time, real-time alarm when a dangerous event or suspicious numerator is found; and intelligent video retrieval technology deals with the storage that has already occurred. Video, which quickly analyzes the video, finds dangerous events, suspicious numerators, and information for each target of interest. The user can then select the event of interest or define the target attribute of interest. Ability to quickly find events or goals that users care about.

    In general, intelligent video surveillance includes features such as perimeter detection, cross-line detection, stay-in-use detection, loss detection, legacy detection, rapid motion detection, fight detection, trail detection, crowd gathering, fire smoke detection, and PTZ target tracking. Video fault analysis, video storage, and playback.

    For different users, the demand for the above functions will be emphasized. Among the above techniques, the methods used for perimeter detection, cross-line detection, stay-in-use detection, missing detection, legacy detection, rapid-movement detection, fight detection, and tail detection mainly include the use of background modeling. Foreground extraction (ForegroundExtraction) extracts the moving target, then uses the target matching tracking technology to obtain the target trajectory, and obtains the target's movement direction, location, and the relationship among the targets. Finally, the above abnormal behavior is obtained according to the set rules.

    Among them, for complex backgrounds, methods for detecting the remains and missing objects in large flow areas can be performed using a special method based on time series regional motion analysis without obtaining the above target detection and tracking techniques.

    The intelligent video retrieval first needs to use intelligent video surveillance detection technology to detect abnormal events. Further, intelligent video retrieval also needs to be based on the detection and tracking of moving targets to obtain objects such as human faces, colors, speeds, and quantities of people and vehicles. And other information. In this way, when an intelligent video search is performed, an abnormal event can be retrieved on the one hand, and the search can also be performed through the appearance end time, color, speed, number, and face information of the target.

    In addition, the system can also give a time and space distribution map of events and targets, allowing users to find the time and events they are interested in. For thousands of monitoring terminals that are now moving, they must rely on intelligent video retrieval technology to find out the events and targets they care about.


    Video surveillance technology is an emerging application direction in the field of machine vision and has attracted much attention. It is also the crystallization of multidisciplinary technologies such as computer science, machine vision, image engineering, pattern recognition and artificial intelligence.

    It can be imagined that when machine vision and image processing technologies were added, the original limitations were broken and a real-time video surveillance system was designed. While implementing video surveillance, the system adds video change detection and automatic video recording capabilities through the use of machine vision technology. The system can automatically identify scene changes, detect moving targets and lock them, and issue warnings and activate storage devices. This can not only save a lot of storage space, improve monitoring storage efficiency, reduce unnecessary playback, and data more targeted.

                                                              Responsible editor: karl


【Return to the front page】
Prev:[depth/attitude] smart manufacturing "eye" machine vision next gold outlet
Relevant articles
Recommended Products
Contact Us

Address: 11th Floor, Kongza Industrial Park, No. 132, Gangzha Street, Fuyong Industrial Zone, Xinqiao, Shajing Town, Bao'an District, Shenzhen City, Guangdong Province, China.

Phone: 86 755 29927385

Fax: 86 755 81466627


Sweep more exciting
Copyright © 2019. Shenzhen Sipotek Technology Co., Ltd. .All rights reserved.粤ICP备05044648号-2