Object Recognition And Cooperative Tracking

Modern intelligent video surveillance system is required to be “better,faster, and more economical”. One of the common requirements is that thesystem can automatically recognize and track the object in the region ofinterest (ROI), and can continuously capture the clear image of the object.However, it seems impossible for the traditional single wide-anglemonitoring system to achieve this goal.As for the discussion above, this paper describes a wide-angle cameraand a PTZ camera that are chosen to construct the object recognition andcooperative tracking system. This system, first uses a confidence basedobject recognition and detection algorithm to describe the confidence ofthe object. Then, based on recognition result, a robust tracking algorithmnamed Tracking-Learning-Detection (TLD) is adopted to do the objecttracking. Afterward, according to the calibration table which is calibratedpreviously between the wide-angle camera and PTZ camera, these twocameras can cooperatively track the object. Eventually, a video objectrecognition and cooperative tracking system is designed and realized. Themain contents of this paper are listed below.1) As for the difficulty of interested object recognition in the videosurveillance system, a confidence based object detection and recognitionalgorithm is proposed in this paper, which is applicable to the videosurveillance system. In this algorithm, the confidence of the detectedobject related to the interest object is under close analysis. Then, the objectof the highest confidence will be selected to be the most interested objectand perform later tasks.2) The TLD algorithm is a long-term and effective tracking algorithm. Due to its robustness, the TLD algorithm is used in the proposed algorithm.Because its real time performance can meet the requirement of this system,this algorithm is adopted to continuously track the recognition result.3) Through the structure analysis of wide-angle camera and PTZcamera, the look up table (LUT) is adopted to calibrate these two cameras.Then, the control strategy of PTZ camera is explored. The proportional anddifferential (PD) control combined with interpolation control is consideredas the ultimate control strategy in this system.4) As for the real application of this system, several applicablehardware is used and corresponding software is compiled to meet thefunctional requirement of the whole system. Eventually, an intelligentvideo surveillance system of video object recognition and cooperativetracking is completed.