object_tracking_in_image_sequences_using_point_feature使用点特征跟踪图像序列中的目标.docx

PAGE 2 Object Tracking in Image Sequences using Point Features P. Tissainayagam and D. Suter raj@.au and d.suter@.au Dept. of Electrical and Computer Systems Engineering Monash University, Clayton, Vic. 3168, Australia Abstract This paper presents an object tracking technique based on the Bayesian Multiple Hypothesis Tracking (MHT) approach. Two algorithms, both based on the MHT technique are combined to generate an object tracker. The first MHT algorithm is employed for contour segmentation (based on an edge map). The second MHT algorithm is used in the temporal tracking of a selected object from the initial frame. An object is represented by key feature points that are extracted from it. The key points (mostly corner points) are detected using information obtained from the edge map. These key points are then tracked through the sequence. To confirm the correctness of the tracked key points, the location of the key points on the trajectory are verified against the segmented object identified in each frame. The results show that the tracker proposed can successfully track simple identifiable objects through an image sequence. Key words: Object tracking, Key points, Multiple Hypothesis Tracking, Contour segmentation, Edge grouping. 1 Introduction The primary purpose of this paper is to track a selected object (as opposed to a single point feature) from the initial frame through the image sequence. The process is an attempt to extend the point feature tracking introduced in [13, 14] to object tracking. In this case, key points from the object are selected using a curvature scale space technique [11] to represent that object. The key points are temporally tracked and are validated against the object contour (obtained by grouping edge segments) in each frame. The tracking technique involves applying the MHT algorithm in two stages: The first stage is for contour grouping (object identification based on segmented edges) and the second stage is