Research On Human Walking Feature Recognition

Traditional identification methods have problems such as high image resolution requirements,low recognition accuracy,etc.The human walking feature combines innate physiological characteristics with acquired behavioral characteristics,and has the characteristics of high individual differences,easiless to camouflage and hide.Therefore,human walking features for identification has a higher recognition rate.Aiming at the problem that the traditional edge detection operator can't extract the continuous sharp contour edge,in this paper after image smoothing,Gaussian background modeling,background subtraction binarization and morphological closing operation,the contour edge of human body is extracted by serial segmentation algorithm.The experiment proves that the contour image extracted by the method is complete and clear,which lays a foundation for human walking feature extraction.In order to extract accurate human walking characteristic parameters,this paper extracts contour centroid key points and the movement characteristics of lower limbs.The contour height,the centroid to the key point distance vector,the lower limb joint angle and the foot swing angle are extracted during the walking cycle,which provides reliable characteristic parameters for identification.In order to improve the recognition rate,an identification model based on BP neural network is designed and the feature classification experiments are carried out.The experimental results show that the recognition rate based on contour centroid key points and lower limb joint motion characteristics is 88.00% and 88.83%,respectively.The recognition rate after feature fusion is 95.30%.The human walking characteristics extracted in this paper can accurately and effectively perform different identities.