Research On Identity Recognition Based On Cross-View Gait Analysis

Computer vision-based automatic human identification is an important research area.An important application is monitoring systems in public spaces to improve security.Computer vision specifically identifies a person by using biometrics that are unique to people and distinguish them from others.The most common methods are detection and recognition using face,iris or fingerprint information.These methods work well in many applications,but there are still some disadvantages.In the monitoring scene,when people are identified,the distance is often far,and most of the features are difficult to obtain clearly and completely.Gait is a posture that people show when they walk,and it is an emerging biological feature.Since the gait is a biological feature that can be obtained from a long distance and is difficult to hide and camouflage,it can solve the problem that the individual cannot be identified at a long distance due to insufficient accuracy of the camera.Gait recognition technology has become an important research direction in computer vision and pattern recognition,and is suitable for intelligent video surveillance systems in large space buildings.Currently,surveillance equipment plays an important role in monitoring high-security areas including banks,airports,military bases,and train stations.However,gait recognition is a challenging task.A person's natural walking style and condition are affected by many factors.Wearing different jackets,carrying items,observing from different perspectives or different environments will bring additional challenges to gait recognition.The premise of gait recognition is to extract the moving target from the video,and the difficulty of extracting the moving target is affected by the background.This article first introduces three methods commonly used in moving target detection:background difference method,frame difference method and optical flow method.Use these three methods to detect moving targets,analyze the advantages and disadvantages of the three methods,and pave the way for subsequent gait recognition.Unlike other biological characteristics,gait is a unique biological characteristic with periodicity.Therefore,gait recognition is also based on periodicity,which requires extracting the complete cycle of gait first,and then describing and extracting changes within the cycle.There are two traditional recognition methods,one is to compress the gait silhouette into an image or a gait recognition template,and the other is to extract features directly from the gait contour sequence.Different from the first two methods,this paper uses the latest concept of gait contour set to extract features based on deep convolutional networks,uses horizontal pyramid matching to extract local features.This experiment was performed in the database CASIA-B.Compared with other methods,it has a higher recognition rate,and has a better effect in the case of different perspectives.Each subject in the gait library usually contains multiple observation angles and multiple walking situations.Coupled with a large deep network,it usually causes slow operation and large memory consumption.Therefore,the second model established in this paper is to use mutual learning networks and improve horizontal pyramid matching to extract local features from both horizontal and vertical directions.This experiment was carried out in the gait libraries CASIA-B and OU-ISIR,and it was confirmed that this method has certain advantages in gait cross-view problems compared with other methods,and can also have certain effects in the case of non-complete cycles.