Research On Target Recognition Technique Based On Gait

In recent years,railway construction in our country continues to improve,the video surveillance of railway is also developing toward automation and intelligence.One of the main tasks of the railway video surveillance and early warning system is to detect the existence of dangerous people or suspects in the surveillance video timely.With the railway mileage and passenger traffic ever-increasing,railway passenger stations have increasingly become high-density personnel distribution centers,and become important sites for criminals to abscond and even terrorist attacks.At present,the surveillance and early-warning of railway system in our country is mainly based on identify verification and artificial surveillance,which has larger limitation.The Gait recognition has received extensive attention due to its unique advantages of long distance,concealed recognition and difficult camouflage.With the appearance of deep learning technique and mature applications in the field of image processing,there has been a new breakthrough in Gait recognition research.There is great significance that Gait identification and early warning based on surveillance video for detection of suspect targets timely.Based on deep learning technique,this paper uses VGG16 deep neural network to study the Gait recognition method,design and develop the monitoring and early warning experimental platform.The main research contents and related optimizations are summarized as follows:First,Gait image preprocessing and Gait cycle detection.Analyzing the foreground extraction algorithm and Gait image characteristics of the current mainstream,a background difference method based on Gaussian model is proposed to achieve efficient extraction of Gait foreground images.And morphological processing and silhouette elimination operations are performed on the extracted images,improving the quality of the Gait foreground image.After the Gait foreground image is obtained,the detection of the Gait cycle is completed through the human body profile high-width information,which avoids calling of a large number of complicated algorithms and improves the utilization rate of the Gait foreground image.Then,for the problem of Gait feature extraction and classification recognition,the gait feature extraction based on VGG16 deep neural network,the generation of Gait candidate regions and the classification and recognition of Gait features are mainly studied.The optimization of the activation function of the convolution subnetwork and the use of anchor generation and non-maximal suppression in the RPN subnetwork is optimized and adjusted,which makes the feature extraction and classification more efficient and effective.Next,according to the implementation process of Gait recognition method,the design of Gait recognition early warning experimental platform is completed.The existing equipments are used to build the Gait recognition platform,completing the relevant environment,software and hardware configuration.In addition,in order to facilitate the integration of Gait recognition function of each part,improve the interaction and visualization of the operation,the design and development of the interactive interface of the Gait recognition and early warning experiment platform is carried out.Finally,the Gait model training and result analysis are based on some of the Gait videos in the CASIA Dataset B database and the actual Gait video.The representative test samples was used to test the Gait recognition early warning experimental platform and Gait recognition method,which futher validated the stability of Gait recognition and early warning experimental platform under certain conditions and the accuracy and feasibility of the Gait recognition method.