Research On Silhouette Based Gait Recognition Technology

As an emerging biometric technology,gait recognition has received widely attention from researchers in recent years.Because it has the advantage of can be acquired at a distance,can be recognized at low resolution,and does not need cooperation etc.,it becomes the only mean of identification in many cases.However,because gait is vulnerable to various factors,gait biometrics is still in its infancy.This paper starts from the research on silhouette based gait recognition technology,focuses on the impact of preprocessing technology on gait recognition.The main work and innovations of this paper include:(1)To solve the problem that the state-of-the-art blind motion deblurring algorithms work well for daytime images,but are less effective for nighttime images with saturation regions,this paper proposed an algorithm with L0 intensity and gradient prior,which is not only effective for the daytime images,but also effective for nighttime images with saturated regions.(2)As the existing alignment way that take use of the body center or upper body center is vulnerable to a variety of factors,which will lead to the silhouettes of gallery and probe are not aligned and will reduce the gait recognition accuracy drastically,this paper proposed to align the silhouettes based on head.The experiments on the popular databases CASIA B,CASIA C and OU-ISIR Treadmill B show that,the gait recognition based on head alignment can greatly improve the accuracy of recognition,especially in the cases that under the influence of backpacks and coats.In addition,to address the problem that no matter what way of alignment is used,there always misalignment because the effect of various factors,the paper proposed a template-fusion solution,experiment results show that the proposed head based alignment along with the template-fusion scheme can greatly improve the accuracy of existing algorithms,and the recognition rate on CASIA B is 15.24 percent higher than the best algorithm right now.(3)Finally,the existing algorithms are all evaluated on the databases that are under strict control of various factors,which will not reflect their recognition performance in real application scenarios.To address this problem,an unconstrained database under real environment was built,and the gait recognition algorithms based on GEI,GEnI and GFI was evaluated on this database.