Gait Recognition Using The Information Of Leg Angles

The physiological or behavioral characteristics of people can be utilized to authenticate their identities. Gait recognition is a relatively new kind of technology, which aims at recognizing people by the way they walk. In fact, gait recognition has gained extensive interest, especially as a result of the increasing demand for intelligent visual surveillance and monitoring systems in security-sensitive environment.Undoubtedly, gait is a quite potential biometric. However, relevant study mainly stays at in the phase of theory research at present, and there are still a lot of urgent problems needed to be resolved, e.g. when the gait frames are under complex background or have occlusion phenomenon, the recognition rates achieved by existing methods are always very poor. To solve this problem, a new gait recognition method is proposed in this paper, which mainly covers the following issues:â‘ In some cases, especially when there is low contrast in gray between the background and moving object, the recovered-object is often made up of holes and distorted parts. To improve the performance of extraction in such situation, a modified method based on the principle of frames subtraction is presented in this paper. Firstly pre-filtering is needed to alleviate the Gauss noise. Secondly, a maximum variance ratio threshold value is used to remove the remaining noise and background. Then some frames are fused to obtain more information about the moving object, and an area for reference is defined at the same time. Finally, the moving object is recovered after scanning the original image.â‘¡Based on an intuitive thought the sufficient individual identity information can be reflected by the joint angle trajectory of body parts during the walking,a gait recognition method using the temporal information of leg angles is proposed. First the pendulum model is referred to for extracting leg parts, and the least-square method is used for boundary fitting, to obtain the temporal information of angle about thigh and shin. Then, the temporal angle information gathered is expanded in the Fourier series form, in consideration of walking activity's periodic property. The genetic algorithm is adopted to search for the harmonic coefficients, which are normalized to produce the gait signature.â‘¢The NN, KNN and ENN classifiers are all applied sequentially to do the sorting task for the four different gait modes on CMU database. Meanwhile, The ROS and ROC index are both used to evaluate the recognition performance of our method, based on the ENN rule. In addition, the possibility of enhancing, the recognition rate by merging the angles information is also investigated.Experimental results on CMU database show that this method can achieve satisfactory recognition rates, even in case of complex background and occlusion; Moreover, the recognition rates can usually be further promoted by merging the angles information.