Research On Sorting, Tracking And Behavior Analysis Of Video Object

Video based monitoring system are used in intelligent transportation field, safeguard surveillance filed and so on more and more, and the intelligentizing research for video monitoring system as a important development field is in the ascendant now. This paper aims at the researching and exploring on the core techniques for realizing the intelligentizing of video monitoring system, including vision based objects sorting, tracking algorithm, objects'behavior analysis and visual attention mechanism mainly. Based on the researching, innovating and practicing of core techniques, a multi-monitor area based universal intelligent video tracking and monitoring system are designed.This paper researches the video images'preprocess technics such as segmenting, feature extracting and objects sorting first. As for the feature extracting of video objects, a color imformation based ICM signature and a SRIC(Silhouette Rotate Interceptive Curve) as the feature of video object are proposed and extracted, then a more distinguishable image signature based on ICM combined with SRIC is realized and achieving advantages complement. As for the sorting of video object, a three OAA(one against all) SVMs based indeterminate class multi-sorting algorithm is proposed, which draw on the experience of the express method used color system, thereby supports the inserting of new class. The experimentation result shows the validity of proposed algorithm.To segmenting and tracking the objects in the video real-timely, after introduced the tMHI(timed Motion History Image) method and improved on its defects to get isMHI(increased-step Motion History Image), a isMHI based multi moving objects tracking algorithm is proposed, which using the layered step-down grey value silhouette to segment and mark the existent sub-motion areas to realize the real-time segmentation of motion objects in video image. Then all motion areas in ever isMHI frame are continually associated with motion objects in video scene to realize the tracking of motion objects'track. For tracking the objects which came to stop after entering the video scene at the same time, an isMHI combines with CamShift multi video objects tracking algorithm is proposed. By this algorithm, when the objects which entering the video scene are moving, the method tracks these objects based on the isMHI, and when some of the objects turn to stay, the method locks and monitors these resting objects based on CamShift instead, so the continue tracking of multi moving objects or resting objects in video scene, and the obtaining of global moving tracks of objects can be realizing. Experimentation shows that the proposed tracking algorithm can segmenting and tracking the multi objects in a high detection rate effectively and robustly, and due to its fast processing speed, the algorithm satisfies the real-time needing.In the research on farther intelligent processing based on tracking result, a HMM based method used for statistic, analysis, and identifying the object's motion behavior is proposed. Take the objects appear in traffic video as example, set the track sequence of tracked traffic objects in video scene as input, and confirm HMM probability distributing models by actual purpose, the behavior identifying, behavior probability statistic and hidden state behavior analysis can be done based on the solving of evaluation problem, decoding problem and learning problem of HMM. To further enhance the intelligence level of vision system, a visual attention mechanism is inducted, and five behavior probability driven object attention models are proposed. By distinguishing the probability of video object doing certain behavior or small probability behavior, the algorithm decides the next processing to take the corresponding steps to the attended object. Experimentation testified the validity of using the purposed methods to solve the behavior identifying problem and visual attention problem.As the realization and application of all the purposed algorithms, at last this paper gives a multi areas based all-purpose system for monitoring and analysis content of video scenes intelligently. This system includes a purposing and realizing of a subsystem, which shooting tracked objects by combining video with digital. In the ordinary facility environment, the subsystem combines the industry camera's real-time property with the digital camera's higher imaging property, and realizing high definition shooting by the network and built-in computer control system, to make it possible to acquire more precise feature data. Otherwise, the main function modules of purposed system such as continue tracking, behavior analysis, feature extracting, object sorting, identification and visual attention mechanism framework are introduced also, and these modules are based on the purposed algorithms in this paper respectively. At the end, the possible applicable situations and corresponding schemes of purposed system are given, and the cooperating performance of the programmed archetypal system is tested, its practicability is validated.