Font Size: a A A

The Front View Gait Recognition Research Based On Multi-feature Fusion

Posted on:2016-09-14Degree:MasterType:Thesis
Country:ChinaCandidate:Y ChangFull Text:PDF
GTID:2308330479950508Subject:Instrumentation engineering
Abstract/Summary:PDF Full Text Request
Gait recognition technology is a biological recognition technology developed in recent years. The main content of the research is to identify the human identity according to the passenger’s walking posture. Compared with other biological recognition technology at the present stage, it has characteristics of long distance, non-contact, hard to fake and imitation and easy acquisition. It has a very broad application prospects in intelligent surveillance and human behavior detection. Therefore, it has great realistic significance to carry out research on gait recognition method. In the existing method of gait recognition, the most common to see is based on the side view, but the research on recognition based on front view is not much. As in gait recognition, considering the direction of travel is an essential factor, it has become a challenging research issues and has great research value. Therefore, in this paper, we take the human gait in front view as the research object, to do the gait recognition.Firstly, According to the characteristics of frontal gait video, the moving foreground always exist in the video image, we cannot use no moving target background subtraction operation, the traditional background subtraction method is improved. Use a last complete frame of the video image as the static background of iteration, performing the background subtraction operation,to obtain a complete moving object region. Aiming at the situation that key information of lower of limbs in gait recognition is easily lost, sub-regional post-processing method is put forward. When connectivity analysis was carried out on the lower part, judge to determine whether a number of connected domain pixel is characteristic of lower limb, in eliminate noise and fill the movement target area empty at the same time, to avoid the phenomenon of the loss of the legs features, to provide more useful information for the positive gait recognition.Secondly, for the positive perspective image sequence, we take the gait feature which changed obviously under a third area of the human body as an interested area for gait cycle detection and feature extraction at first, in order to improve the processing speed of recognition system. Then using symmetry of the width of the leg in interested area, to get the line to divided into left and right leg, to avoid the hand shaking when human body target moving. Then put forward two kinds of new method suitable for positive gait cycle detection that using the lower limb length ratio of human walking and lower limb swing area to calculate the gait cycle.Thirdly, in view of the present positive gait recognition rate is not high, the gait feature is too single, respectively in this paper, based on the idea of feature fusion to extract the body interested area principal component analysis(PCA) as static characteristics, extraction of lower limb length ratio and swinging area as a dynamic characteristics. Using the theory of data fusion, the static and dynamic characteristics with fusion, and make up for the shortcomings that single feature information is not complete.Finally, Using support vector machine(SVM) as classifier, and training and recognition on the CASIA gait database provided by the Chinese Academy of Sciences. The experimental results show that, using the fusion feature classification recognition method has achieved a higher recognition rate, make up for the problem of low rate by using single feature recognition, it has certain feasibility.
Keywords/Search Tags:gait recognition, lower limbs length ratio, feature fusion, Support vector machine
PDF Full Text Request
Related items