Research And Implementation Of Multiple Objects Tracking Based On ICP Algorithm In Visual Surveillance

Due to the constant social economic growth, an increasing number of the surveillance equipment are widely used in security surveillance field, resulting a considerable increment in the need of the intelligent surveillance system. Intelligent surveillance system, based on the computer vision technology, has ability to recognize and track the object from the video frame. Furthermore, some of them can achieve high-level functions, such as understanding and analyzing the behavior of the object, forecasting abnormal actions and alerting. Therefore, with the smart home devices becoming general, intelligent surveillance system will step into any places in the need of security. However, limited by algorithms and computer performance, designing a robust real-time surveillance system is still a hot topic in this field. Focusing on the difficulties of the surveillance system, such as object detection, object tracking, light variations and object deformation, a series of research has been carried out. The main work of this paper lists as follow.ⅰ. To detect targets from frames with background noise, this paper presents a method based on the background subtraction and ICP algorithm. This method fully uses the characteristic that surveillance video’s background is constant. At the beginning of the tracking, our approach extract constant background image of the surveillance area from the video frames. Then difference between the background image and current image will be taken. Further, the difference image is converted to point cloud data, and the object detection is achieved by using the ICP algorithm to match point cloud data with target models that made in advance.ⅱ. Implement object tracking by using Kalman filter algorithm. Using the advantage that Kalman filter algorithm can predict the location of object, the paper design multi-target tracker to predict the object’s location in the next frame and detecting targets around this location using the ICP algorithm. In this way, the object tracking will be achieved accurately and rapidly,ⅲ. System reply dissimilar morphology of different object and changes of object’s shape with multi-model approach. With an adequate consideration of the diversity of the object in surveillance area and changes of targets’shape, system extracts various models of each object, tracking multi-object with multi-model.iv. System design some effective methods to process problems. System continuously extracts and updates the background, started by timing or manual. System have design excellent method to process covering between close targets and choosing the best model from multi-model.The paper is focused on the realization of tracking object in visual surveillance. The main innovation of this thesis is that detecting targets by matching multi-model with multi-target using ICP algorithm, building Kalman filter tracker that tracking multi-target accurately and rapidly. Not only The methods of updating background and matching multi-model to multi-target, but also the solutions of solving covering and targets getting in or out are certain innovative. Moreover, converting theoretical study into robust real-time tracking system is of certain innovation.