Research On Carrier Detection Based On Gait Recognition

Carrying object detection and gait recognition are hot issues in the field of computer vision.The existing methods face the following two prominent problems:(1)Although the video surveillance system has been widely used,it is often prone to high-risk terrorist attacks or public security incidents in places where monitoring of dead ends or inappropriate monitoring systems is set up,such as intentional indwelling of dangerous goods or robbery.With the change of the carrier of the suspect,how to use the surveillance image outside the place where the terrorist or security incident occurred to carry out the human identification and the change detection of the carrier,and then the early warning or the follow-up is an urgent problem to be solved;(2)The existing method uses the gait energy image as the input of the model in the gait recognition.In the synthesis process of the gait energy image,the non-uniform situation of the person region identified by the target detection is not considered,so that the correct center of gravity coordinates cannot be found,and the inaccurate gait energy image is synthesized,resulting in low gait recognition accuracy.In response to the above problems,the main work and innovations of this paper are as follows:(1)The synthetic part of the gait energy image.Aiming at the problem of finding the correct center of gravity for some non-connected people's regional images,this paper proposes a gait energy image optimization synthesis algorithm based on center of gravity alignment.First,the conditions for judging the break and the headless condition are obtained by learning,and the images are distinguished.For the broken image,the head is merged with the torso,and the headless image is found and deleted;then,the retained image is calculated.Accurate barycentric coordinates;finally,the gait energy image is synthesized using the center of gravity alignment technique.Experiments show that using the algorithm to align the post-synthesized gait energy image,a better synthesis effect is achieved,and the accuracy of gait recognition is improved.(2)The double recognition part of carrying object detection and gait recognition.When it is impossible to film the occurrence of suspicious behaviors such as theft or placing dangerous goods,it is necessary to make a correct judgment of the danger that has occurred or will occur in a timely manner.If the neural network is used to judge whether the same person is carrying a change or not,although it can meet the requirements,since two networks are used,a large number of parameters need to be calculated.For this problem is proposed based on improved twin network of gait recognition and object detection algorithm,to carry through the traditional twin neural network was improved,make it a double output structure,compare in and out of the two video of a particular place,not only can judge whether it is the same man,but also to determine whether there is a before and after carrying state changes.In training,the algorithm will be generated by two video gait energy image of two pairs,and defines each pair of double label,so,this method does not need to be directly observed with content change process of transfer,also need not establish contacts tracks data in advance.Compared with the traditional gait recognition algorithm,the accuracy of gait recognition is also effectively improved.This method gives the same person(gait recognition)and whether to carry the change(portable detection)at the same time.The accuracy of the judgment reached 87.54%.Using a network to judge two problems at the same time not only ensures the recognition accuracy,but also saves the calculation cost of parameters.