Research And Implementation Of Single-person Gait Recognition Based On Video

Gait recognition judges the identity by analyzing the posture and action of pedestrian walking sequence.Compared with other biological characteristics,gait is not affected by the low resolution and distance of the image,does not need the active participation of the target,and is difficult to be glossed over.However,the specific application effect of the traditional gait recognition method is still affected by the external factors such as pedestrian's clothing,backpack and so on.To solve this problem,this paper combines two kinds of gait recognition methods based on silhouette sequence and skeleton sequence,then studies and analyzes their implementation methods and specific applications.Firstly,the gait silhouette obtained by the traditional background difference method have the shortcomings such as quality,practicability and robustness.Aiming at this,this paper designs a gait silhouette sequence recognition method based on attribute information guidance.In this method,human body instance segmentation technology is used to obtain more accurate gait silhouettes.This can improve the running speed and universality in the actual scene.On this basis,the attribute classification task is added to recognition network.It can enhance the recognition ability of the model for complex states such as clothing and backpack.In addition,multi-scale time feature extraction module is introduced.This makes the model integrate different time scale information to form a more discriminative space-time feature vector representation.Then,body parts of gait silhouettes have adhesion,occlusion and big difference with the normal scene in different clothing and backpack scenes.Aiming at these problems,this paper uses human skeleton information to identify.The core idea is to extract the key points of human skeleton in the original image sequence based on deep learning,and then construct the gait skeleton map structure to assist the subsequent gait recognition.In the specific implementation,this paper introduces several same convolution modules of spatiotemporal graph.This makes the extracted gait skeleton sequence fully integrate the feature information of both spatial dimension and time dimension.Finally,on the basis of the above two works,this paper designs a gait recognition method based on multi-source information fusion of gait silhouette and gait skeleton graph.This method combines the spatiotemporal feature vectors of gait silhouette and gait skeleton,and then constructs a more effective pedestrian gait synthesis feature vector.The extracted gait feature vector is used to measure the similarity with weight in gait feature database,and then their identities are identified.The experimental results on CASIA-B gait dataset show that,compared with gait recognition only based on silhouette sequence or skeleton sequence,synthesis of two kinds of information can achieve a greater degree of accuracy improvement.In conclusion,our method based on multi-source information fusion can make full use of the two types of gait information to supplement and promote,which provides an effective improvement direction for subsequent research.