Study And Application Of Pedestrian Recognition Based On Gait

Due to the rapid development of artificial intelligence,the research of graphics has undergone a qualitative leap,and various identification methods have been widely used in various fields.Many heavy industry enterprises,such as steel,coal mines,and electric power industries need to identify personnel.The video surveillance system has been introduced as an auxiliary tool for production safety.While the system is being identified,it can also monitor key scenarios,which greatly facilitates the supervision of enterprises.However,most enterprises still rely on human eye observation to complete the monitoring task.However,since mines or power plants usually need 24 hours of continuous operation,it is difficult for workers responsible for watching and monitoring to concentrate for a long time,and it is inevitable that there will be leakage.Therefore,manpower is obviously not very reliable.In order to solve this problem,some companies have introduced face-based identification systems.However,due to the work wearing a safety helmet,and the worker's face is often stained with dirt,the facial recognition system does not perform well.In recent years,gait recognition has received more and more attention as an emerging biometric technology.Compared with other biometric technologies,gait features are not affected by appearance pollution and are easy to implement remote noncontact.The advantages of identification are suitable for complex industrial production environments.At present,gait recognition still has problems such as difficult feature construction and low cross-view recognition rate.This paper proposes a pedestrian recognition method based on gait energy map and a pedestrian recognition method based on gait sequence.At the same time,the recognition rate is more robust to the change of viewing angle and appearance.The research content of the thesis mainly includes the following aspects:(1)Improved foreground detection algorithmAccording to the complex characteristics of downhole light,an improved hybrid Gaussian model algorithm is presented.The algorithm uses pixel gradient instead of pixel intensity as the input of the mixed Gaussian model to solve the problem of foreground detection error when the light is abrupt.The gradient in four directions is used as the basis for discrimination,which further improves the accuracy of foreground recognition.(2)Pedestrian recognition method based on gait energy mapA neural network structure is designed to extract multi-dimensional gait energy map features,and use sub-network to learn the weight of each feature to improve feature validity and optional invariance.In addition,for the ternary loss function The structure and training methods are optimized to speed up the convergence of the model without affecting the accuracy.(3)Pedestrian identification method based on gait sequenceA neural network structure is designed to integrate single-frame features into the same space,which reduces the dimension of features.In addition,the network can extract the spatio-temporal features of gait movement and integrate with single-frame features to improve the wearing.Variable anti-interference ability.Using gait characteristics to complete pedestrian identification in industrial scenes can provide guarantee for the safe production of enterprises and facilitate the supervision of enterprises.At the same time,this is also in line with the development trend of industrial intelligence,which is of great research significance.