Unmanned Following Car Based On Gait Recognition

With the development of science and technology,the intelligentization of the automobile has become the trend of the industry.Unmanned vehicle is gradually moving towards diversification and specialization as an important tool for the study of smart driving.It can effectively sense the information in the surrounding environment,make decision and plan the global and local information,realize the control of the track and so on.Target recognition is a key technology,and it is also the key to realize the function of the unmanned vehicle.Because of the complexity of the surrounding environment and the interference of pedestrians,the unmanned vehicle is not difficult to lose track of the target.As a unique feature of the human body,gait recognition has the characteristics of long-distance,non-contact and imitative,which play an important role in target recognition.This thesis will combine human gait as a target identification feature and fuse it into an unmanned vehicle to achieve human target following.It will deal with the problems of gait following by unmanned vehicles as following.1.The pre-processing of moving targets in gait video.First,the hybrid gauss model method is utilized to separate the moving targets from the background.The target image is then morphologically treated for problems such as holes and noise.Finally,in order to secure the unity of the target contour height,the target is normalized.2.The features of 2D human body contour.Firstly,the key points of 2D human skeleton are extracted by the proportion relation of human skeleton;Secondly,the feature vectors of human body recognition are established based on the features of skeleton length,skeleton swing angle and swing period Finally,the nearest and support vector machine are used to test the recognition effect.3.The processing of human body contour clothing,accessories and contour map time sequence.Firstly,the feature of the Human Body Contour is masked by a convolutional neural network;then the time characteristic of each frame is extracted.Time Information Pool of each contour map is then mapped to the contour period level feature.Finally,gait recognition is completed by the periodic level features to more differentiated space and measuring the similarity between the GAIT CONTOUR sequences.4.By using Turtlebot 2 as the experimental platform,the gait contour and skeletal features of the lower limbs are fused to identify the following effect of the unmanned vehicle.Two kinds of experiments are intended.One is the experiment of many people walking side by side without block,and the other is the experiment of crossing in the course of walking.A timer is introduced to estimate the following statement.This thesis extracts the characteristics of the human lower limb bones,swing angle,period and contour,and designs an unmanned following car based on gait.In this thesis,based on the ROS platform,we deeply researched the gait-based following method.The hardware used can be an ordinary camera,which reduces the equipment cost.In this thesis,the lower limb gait contour fusion technology is proposed,which effectively avoids the interference of clothing and other accessories on the gait,enables the unmanned vehicle to accurately identify the target indoors and outdoors,and ensures the robustness of the recognition algorithm.