A video object segmentation algorithm based on Snake models

A new video object segmentation algorithm based on the improved greedy Snake model is proposed to solve the problem of object tracking. This algorithm combines temporal and spatial information together. Firstly, the video sequence can be divided into segments due to the fact that the movement trend of adjacent frames remains similar in a short period of time, and each segment has k frames; Secondly, the first two frames of each segment are recognized as key frames, and the rough contours of the moving object in the first two frames are acquired automatically by using the motion detection; Thirdly, the improved inter-frame greedy Snake iteration is applied to get the precise contour; Fourthly, the intra-frame moving vectors of the moving object centers in the key frames are used to predicate the initial contours of the moving object in the subsequent frames; Fifthly, the improved inter-frame greedy Snake iteration is applied for the non-key frames to get the precise contours on the basis of the initial contours, and then the video object segmentation can be realized for all the frames. Compared with the traditional methods, the proposed algorithm overcomes the disadvantages of drawing the initial contour manually. Furthermore, the greedy Snake method in the spatial domain has been improved with high accuracy, speed and many other obvious advantages. Experimental results indicate that the new method realizes the corresponding match of adjacent moving objects and gains accurate segmentation results through the improved greedy Snake method.