The future of video surveillance: multiple camera tracking |
|
Please, note that the article is automatically translated from Russian into English, so the translation may not be accurate. Nikolai Ptitsyn Videoanalytical software has become almost a mandatory component of the modern system of video surveillance. Intelligent Video allows to raise the productivity of safety and reduce the psychological burden of the operators situational centers. Multi-support - the next stage of development videoanalytical technology and the concept of security system as a whole.
A key function of the system for multi-tracking is the automatic registration of the trajectories of physical objects in the vast area controlled by multiple cameras. If, in the single as possible accompanied by the independent data processing for each channel, the multi-chamber system must analyze all of the channels integrated. The multi-system allows three-dimensional geometry of space and positioning of cameras, as well as making rational assumptions about the most likely trajectory of motion of an object, even if he is temporarily out of the joint zone control of all cameras. The task becomes much more complicated when the objects are accompanied by numerous and poorly distinguishable, or, conversely, poorly comparable because of its variability, different angles of observation and image recording conditions. Thus, it is about creating an artificial intelligence capable of simultaneously carrying tracking of multiple objects across multiple cameras. In contrast to the single-chamber tracking this problem immediately beyond the capacity of the human brain. Technology for multi-tracking is primarily in demand in sectors such as health, safety, transportation, marketing, retail sales and interactive advertising. Fig. 5.1 illustrates the Multi-support of a chosen subject in the main hall of the Minsk railway station.
Consider a typical scenario of multi-chamber system in the field of mass stay of people. Interactive mode
Fig. 5. Return of object 1548
in view of the first chamber Assume that the security officer drew attention to the suspicious behavior of the subject in the passport control area of the international airport. The employee selects the subject on the screen that displays video from one of the cameras, and instantly receives the trajectory of this character in three-dimensional model of the building.Shows the trajectory unauthorized access through the service space. Prompt arrest the offender is also produced with the support tracking system. Automatic modeSignal suspicious behavior can be automatically formulated based on the rules of the regular model of the person or group on the controlled object with the current time and mode of operation. Examples of simple rules are:
Analysis of the statistic dataRules can be chosen optimally based on statistical data on the movement of people gathered at any time interval. Perhaps fully automatic learning system "pattern" of conduct for the subsequent discovery of unusual movements. For the transportation industry multi-chamber system can track the behavior of passengers and to accurately measure load in each direction at different sites. In London and Stockholm recognition registration plates are used to control the fees for entry into the city, and if necessary, may produce Multi-escort cars in the city or along the route for security services. In marketing and retail sales of multi-chamber system enables us to investigate the behavior of shoppers in malls and supermarkets. In the case of system integration and maintenance of broadcast advertising on the digital panel displays opportunity to evaluate the effect of advertising on individual customers, as well as to show a visitor related videos in the process of being relocated to the mall.The modern level of technologyA number of western universities have created a system for multi-skilled maintenance and have demonstrated their efficiency in their campuses. Development consists of several cameras and a workstation, generating processing stream or recorded video. In the area of the camera are moving freely students. Flux density is low (one to several square meters), but objects are regularly overlap. Results of treatment are the coordinates and trajectory of the people on the two-dimensional plane-controlled area. Experienced system showed good tracking accuracy, it is sufficient to justify the practical value of the development. Commercial introduction of constrained by the following factors:
At the level of twin-tracking problem of segmentation of people in heavy traffic on a different scale. Thus, the algorithms of computer vision are often "wrong" during separate the object from the background (for example, when people are close to each other, partially or completely overlap, cease to move, look outside the box). While a variety of appearance and behavior of man is boundless, a simple deviation from the model (say, moving to a wheelchair, or disclosure by the newspaper) could "enter" computer system "error." People appear at different distances from the camera, therefore, have different parts of the image detail and informative. Often difficult to achieve uniform illumination over large areas. These factors substantially increase the computational complexity of video processing algorithms. Additional uncertainty creates the scene physical barriers that restrict camera review, such as pillars and stalls.If the program is "losing" the object at some point in time, then there is a rupture of the trajectory and the lost opportunity to trace the movement of the object from the start and end points. At the level of support for multi-central technical problem is to ensure the invariance (constancy) of grounds on which the objects being compared after the temporary disappearance of the zone control of all cameras Accompanied by an object (a person) can be observed at different angles, at different distances and in any condition (sitting, standing, going to be running). Different types of lighting (natural, artificial) impede the use of color features, since they depend strongly on the emission spectrum of the illuminator. These factors lead to instability of the numerical values of attributes and comparing objects to errors in the transition from the zone of one cell to another. Accurate calibration of cameras in three-dimensional object under control improves system efficiency for multi-tracking, especially if the monitoring cameras overlap.Ways to overcome technical problems in the multi-tracking system
Assessment of the accuracyTo assess the accuracy of the system for multi-tracking requires special equipment, provide automatic testing, because manual testing is too time-consuming and does not provide a sufficiently good repeatability of experiments. Automated testing requires a canonical set of TV spots, significantly marked by the experts. For example, a MCTS (Multiple-Camera Tracking Scenario) developed the scientific division of the Ministry of Internal Affairs of Great Britain and includes 140 hours of video, which was simultaneously recorded with five cameras installed at the international airport. Development of the multi tracking system is an urgent scientific and engineering problems whose solution is now in demand by users of various categories. There is an opportunity to analyze the behavior of many people across the protected object, and not just in the field of view of one camera. As we refine the mathematical theory of computer vision and computational power of hardware support for multi-application systems will expand into new industries. |
