Study On The Key Techniques For Human Identification At A Distance Based On Fusion Of Gait And Face

Biometric recognition and verification, which has gained highly interest from academia and business, will be one of the key techniques in the security system in the future. Gait recognition has been the most potential research topic for its advantages in the filed of long distance human identification. The research of gait recognition is still in the early lab stage. A reliable and robust gait recognition system which can be used in our daily life still needs extensive efforts. Multimodal biometric fusion is a new study trends in the biometric recognition field. In this thesis, the human identification work was done by fusing gait and face information. The key techniques were studied.DatasetB of the CASIA gait database provided by the institute of automation, Chinese academy of sciences was used for the human identification experiments. Firstly, human silhouettes and human faces were extracted automatically from the video monitoring images. Secondly, a 3D human body model was created, which was then used to track the gait sequences to obtain the human joints angles. The tracking results were used to create the model based gait energy images (MGEI) used as the holistic gait feature. Wavelet transform was done to the joints angles of the human legs to form the dynamic gait feature. Thirdly, face image preprocessing algorithms and face recognition algorithms were studied and compared by experiments. Super solution image reconstruction techniques were used to increase the resolution of the face images. Finally, fusion algorithms such as D-S evidence theory, max rule, min rule, sum rule and voting rules were used to fuse the gait and face feature. The experiments showed that the recognition rate was the highest after the fusion of gait and face by D-S evidence theory. The recognition rate is 94.74%.The innovations of the studies were:①A self-adaptive 3D human body model was created. This model could be used to obtain 3D information of the human's joints angles and human's shape by human tracking which could represent the human.②A new gait feaure of MGEI was proposed. The experiments showed that MGEI can be used to identify human③. A new gait dynamic feature based on Wavelet transform was proposed. Low frequency component of the Wavelet transform of the human joints angles were used as the gait dynamic feature. Experiments showed that after Wavelet transform the gait recognition rate was increased by 3.50%.④A Singular Value Disturbance based 2DPCA method was proposed for face recognition. And super resolution image reconstruction techniques of POCS and IBP were used to increase the face recognition rate.⑤A new method of fusing gait and face by D-S evidence theory for human identification at a distance was proposed. The human recognition rate was effectively increased.