A Study Of Gait Feature Extraction With Human Clothing Invariance

The use of video surveillance to identify pedestrians is one of the current research hotspots in the field of computer vision,which is of great significance for ensuring public safety and detecting criminal cases.With the widespread application of convolutional neural network technology in the field of pedestrian recognition,its recognition automation and accuracy have been significantly improved.However,through the research status,it is found that the existing pedestrian identification methods have the following problems: the existing methods rely on the appearance characteristics of the pedestrians,that is,when the pedestrians change clothes,backpacks and other appearance characteristics,the accuracy of these methods is significantly reduced.In order to solve this problem,this paper uses human body reconstruction technology to process the video data to remove the interference caused by changes in appearance features.On this basis,a deep learning framework based on pedestrian gait features is proposed.First of all: This paper proposes a human body appearance correction algorithm based on SMPL 3D human body reconstruction to extract a 3D human body model.This obtains a continuous 3D pose sequence from video clips,and reduces the estimated 3D pose by assuming that the 3D parameters change slowly over time.error.Secondly: Map the three-dimensional posture parameters back to the two-dimensional space,completely remove the information of clothing and carrying objects,and modify the appearance changes caused by different clothing or additional carrying to maintain the invariance of human clothing.Then: A Person re-identification algorithm for video surveillance networks is realized through the mapped image sequence.Finally: the reconstructed image sequence of the three-dimensional human body is represented by a gait energy image(GEI),and a gait recognition algorithm is implemented based on an improved convolutional neural network(CNN).Experiments on public data sets verify the effectiveness of the method proposed in this paper.This paper conducts experiments on the CASIA-B data set,and sets up two different situations for experiments.On the CASI-B data set,compared with the existing Person re-identification methods,the algorithm in this paper works when pedestrians change different clothes.The accuracy of Person re-identification increased by 17.6% on average;while carrying different objects,the accuracy increased by 21.9%on average.Compared with the existing gait recognition algorithm,the algorithm in this paper has an average accuracy of 9.8% higher when pedestrians change different clothes,and an average accuracy of 1% higher when pedestrians carry different objects.The experimental results show that,compared with the existing methods,the method in this paper effectively improves the recognition accuracy of pedestrians in gait recognition and pedestrian re-recognition when changing clothes or carrying objects.