Method And Implementation Of Gait Recognition System

With the development of society and the growth of safety requirements,biometric recognition technology has received widespread attention.Gait is an emerging biometric recognition technology,which performs identity authentication through walking posture.Compared with face recognition,it has the advantage of long-distance non-contact collection.Compared with Person Re-identification,it has the advantage of not being disturbed by clothing color.So,gait recognition has great potential in the field of surveillance and security.However,gait recognition can only be used as an auxiliary means in the actual scene system application.Gait recognition accuracy is affected by some external factors,such as camera perspective changes,pedestrian appearance changes,and poor segmentation of gait data.The above factors bring great challenges to the gait recognition task.In view of the above problems,this thesis studies and designs a gait recognition model to extract more robust contour features and richer time series motion features,and designs practical and real-time engineering applications in the park network monitoring scenario.This thesis studies the gait recognition problem in this real complex scene:1.This thesis propose a multi-scale fusion residual gait recognition network: Res Gait.The Res Gait model includes two parts,the horizontal local feature extraction module(HLFE)and the multi-scale fusion module(MSF).The study found that when a pedestrian walks with a backpack or an object,the local shape and motion features are affected,while changing clothes and viewing angle changes nearly half of the shape and motion features.Therefore,it is believed that using both local and global features should be better.A convolutional residual layer R-Conv is designed,which is composed of horizontal block convolution and ordinary convolution in parallel.The HLFE module is composed of multiple different R-Conv layers.Secondly,when the silhouette of pedestrians changes,it does not affect the time-series motion information of pedestrians.Therefore,it is necessary to extract more refined time-series motion features.The MSF module extracts temporal features from the outputs of the HLFE module at two different scales through two temporal feature aggregators,and finally fuses the features of different scales to obtain the final gait feature.Finally,this thesis conduct sufficient ablation and comparison experiments on CASIA-B,the most widely used public dataset,and the algorithm achieves an average accuracy of 91%.Res Gait outperforms all existing algorithms.2.Further,this thesis take the proposed gait recognition algorithm Res Gait as the core for system engineering design,based on the network monitoring system in the park scene,realized the intelligent gait recognition system in the park.This thesis first analyzed the functional requirements and designed the overall architecture,set up and calibrated cameras,configured switches and servers from a hardware perspective.Deep learning algorithms such as target detection and tracking,human attribute recognition,pedestrian segmentation,gait recognition and matching have been trained from the software level.Then carried out algorithm and engineering optimization to meet the requirements of both accuracy and efficiency.Then developed a database and front-end Web interactive interface,realized the animation visualization of target pedestrian trajectories,human attribute analysis,and accompanying personnel analysis.Finally,a detailed functional test and performance evaluation of the accuracy and efficiency of the algorithm are carried out.The system meets the requirements of real-time and practicability in the intelligent monitoring scene of the park.This thesis proposes a more robust gait recognition algorithm and achieves better performance.On the basis of the proposed algorithm,the gait recognition system engineering application of the park network monitoring scene is realized.At the same time,it is found that there are more challenges in the real scene,and further research and exploration in academia and industry are needed.