Gait Contour Extraction Based On Improved Snake Model

As a new method of the biologic recognition, gait recognition identifies human through the way he/she walks. Compared with the other biologic recognition methods, gait recognition has been regarded by the research institution of many countries, for it has many unique advantages, for example, the target can be recognized in a long distance without contacting the equipment, and this process will not easily become aware of, in addition, it's very hard to change or imitate people's walking habits, etc.In the gait recognition, first and foremost step is to extract the moving object out of the background. Integrated and effective extraction of moving object is very significant to subsequent processing including the target classification, feature extraction, feature expression, and the last recognition, because in the subsequent processing, it is only considered of the image elements which corresponded to region of the moving target in the original image. Under the influence of the complex background, extracted contour is always very rough, even not close, which is obtained form the classical methods, for example, the background subtraction methods, the frame-to-frame subtraction methods, and the methods based on the moving fields. For this reason, moving information is lost and the subsequent processing is influenced badly, and it also is the reason why the gait recognition rate can not be increased in a wide-range. In order to increase the recognition rate, some of the researchers patch the contour through combining computer automatic positioning and manual work, but this work will cost much time. Thus, integrated and effective extraction of moving object has become one of the keys in the gait recognition, and it is also one of the difficulties.This paper researches on an algorithm that is able to extract the gait contour accurately and completely. A number of classical segmentation algorithms have been proposed to solve this problem, but they have one flaw in common, that is the segmented contour is always incomplete because of the complex background. This paper introduces a model called Snake which can patch the contour automatically and accurately, at the same time, the model is improved to decrease the processing time. At last, the paper estimates the validity of the algorithm in three aspects. The performace results of the experiments on the CMU database prove that the gait contour can be completely and closely extracted by the proposed method, and recognition rate is also increased effectively. In order to improve the performance limitation of original Snake model, many researchers have proposed some improved Snake models, for example, the Balloon model, the distance model, the GVF model, and the Greedy algorithm, etc. Among these, GVF model is the most popular one for its good convergence property. This paper analyzes the human body characteristics and chooses the GVF model to search the boundary concavities. But the variational calculus used in GVF model is very slow, this paper combines the GVF model and greedy algorithm to decrease the progressing time.Because different researchers use different databases and different segmentation algorithms, so the estimating methods are not consistent. This paper generalizes the estimating methods, and estimates the algorithm from the numbers of the moving targets, the effective area ratio of the moving target, and the gait recognition rate. The estimating results are given as experiment data. Compared with the background subtraction methods, the algorithm proposed in this paper can obtain better segmentation.