Implementation and Comparative Analysis of Various Motion Segmentation Techniques

Abstract

The moving object detection and tracking in video sequence is current research area in computer vision. Object tracking is a key step for object recognition, navigation systems and surveillance systems. We propose a simple and e ective approach for object tracking in video sequence, based on motion segmentation. The algorithm is described in detail and simulation experiments are performed on object tracking which verify the tracking algorithm's competency. The proposed algorithm consists of four stages i.e. image segmentation, feature extraction as well as object tracking and distance estimation. This algorithm is run on consecutive frames of video se- quence; the extracted features are matched in the next frame and the motion of the object is calculated in both X and Y directions frame by frame. Hence, the moving object in the video sequences will be tracked and distance traversed by the object is estimated. The limitations of this approach include only a single object tracking as well as obscured detection on black coloured object. Even for the sequences contain- ing dynamic background, above approach is not able to distinguish between dynamic background and moving object. To diminish all above difficulties with the proposed approach, another technique based on optical ow is studied from the literature and implemented with some modi cations. The presented approach consists of generation motion vector and motion histogram. Based on the information of consequently updated motion histogram,dynamic back- ground is removed from video sequence and moving object is segmented. Finally morphological operation performed for removing unwanted noise and shaping of seg- mented object. Quantitative evaluation and comparison with the existing methods show that the proposed method provides much improved results.