| With the application of video surveillance in daily life,more and more attention has been paid to using computer vision technology to identify the biological characteristics of images.Gait recognition performs identification by analyzing the walking posture and body structure of the target pedestrian.Compared with the traditional short-range(face or iris recognition)or contact(fingerprint recognition)recognition technology,gait recognition technology has the advantages of longdistance and non-contact,providing strong support for social security,criminal investigation and other industries.In recent years,many researchers have continuously improved the performance of gait recognition,but the recognition accuracy is still affected by many factors,such as changes in visual angle and appearance.In response to the above issues,a new gait recognition model is proposed,which extracts the local short-term features and global long-term features of silhouette images on the gait sequence,and combines them with the real human posture information calculated by monocular vision for gait recognition.The main research contributions include: 1)The frequency of walking motion may change greatly in different scenes,so global long-term attention and local short-term attention are extracted from human silhouette sequence to enrich gait information and improve the accuracy of gait recognition.2)Compared to human silhouettes,the calculation of real posture information(such as real height,step frequency,step length,etc.)will not change significantly in different scenarios.Using only silhouette images can only provide limited observation of contour or motion features,a method for calculating human posture information is proposed,in which static and dynamic posture information is calculated by monocular vision.3)Based on the calculated human pose information,a posture feature extraction network is proposed,which is convenient for the whole model to combine human silhouette features with human posture features for gait recognition.Through the experiments on multiple datasets,the effectiveness of the proposed method is proven.Afterwards,based on the proposed algorithm model,a cross scene gait recognition system was designed and implemented to meet the requirements of intelligent security.This article first analyzes the system functional requirements and designs the overall architecture.In terms of functionality,algorithm models such as instance segmentation,posture estimation,and gait recognition were implemented and integrated,and algorithm and engineering optimizations were carried out to simultaneously meet the accuracy and efficiency requirements of the system.At the same time,the development of databases and front-end pages has been carried out,achieving the function of identifying and matching target pedestrians through gait features.Finally,detailed functional testing and accuracy evaluation were conducted in multiple scenarios to verify the application requirements of the system in cross scenario monitoring scenarios. |
