The Research Of Gait Recognition Based On The Vector Diagrams Of Human Walking Characteristics

Public spaces like metro,train stations and airports are featured with characteristics such as high population density and high passenger flow,which make them major targets for criminous fugitives and even terrorists.The demand for smart visual monitoring technology in these places has been increasing on a daily basis.As a non-contact,non-invasive,difficult-tocamouflage-and-imitate new biological feature that can be remotely monitored,human gait will be playing an irreplaceable role in public transportation safety.Promoted by the rapid development of AI,gait-based identification technology has made new breakthroughs,but there are technological challenges imposed by the change of clothes,carried items,perspective and walking speed and it is unable to process in real-time,all of which render impossible for gaitbased identification and monitoring system to be put into use in massive scale.In allusion to the change of concomitant variables such as clothes and carried items under multiple perspectives,which leads to imperfect identification,the gait recognition based on the vector diagrams of human walking characteristics are proposed.The main works are illustrated as follows:(1)In consideration of the fact that most gait recognition tasks require real-time processing of complicated outdoor situations,the human posture estimation algorithm OpenPose,which uses the underlying network as Mobilenet,is proposed to acquire the vector diagrams of human walking characteristics.First,YOLOV3 is used,in combination with the gait contour aspect ratio,to collect the average gait cycle from datasets,which is used as the setting of time_setup,the hyperparameter of the gait space-time network.Then,the PAFs of heads,which contribute nothing to gait recognition,are removed and data standardized.At last,PAFs are stacked according to time sequence,hence the vector diagrams of human walking characteristics proposed in this paper are formed.Experiments are conducted in some of the CASIA-B datasets,and the results show that the characteristic description acquired in such way not only retains abundant space-time information,but also avoids redundant information that is adverse for the learning of characteristics of space-time network and model training,and it can better address issues of difficult-to-identify resulted from the change of concomitant variables such as clothes and carried items.(2)Issues such as poor robustness and low recognition rate of the algorithm resulted from the change of concomitant variables such as clothes and carried items under multiple perspectives are innovated in allusion to the existing gait characteristics learning and sorting algorithms,and a neural network that can better address the influence of the change of clothes and carried items under multiple perspectives is designed.The vector diagrams of human gait are used as the gait space-time network input,which learns the characteristics of gait space based on residual learning modules,and learns and recognizes networks based on the characteristics of how LSTM learns the characteristics of gait time,and it determines the sorting on softmax layer,which is at the end of the network.(3)The outdoor database of three types of gait statuses under 18 perspectives is created(normal,change of clothes,carried items),and same-state and cross-state experiments under multiple perspectives are conducted in the database,the applicability and robustness of the gait recognition,which is based on the vector diagrams of human walking characteristics,are verified.At last,on the basis of the research on the algorithm,the identification system based on the vector diagrams of human walking characteristics are designed and realized with Tensorflow,OpenCV,and PyQt5,and gait recognition tests are finished with the system in the self-created database,which showed great results,indicating that the proposed method is able to effectively improve the algorithm recognition rate and robustness while considering the change of clothes and carried items under multiple perspectives.