Research On Specific Pedestrian Tracking Technology For Service Robot

Service robots often work in uncertain environments dominated by people,such as exhibition halls,homes,restaurants,etc.It often needs to interact with a designated pedestrian in a large number of pedestrians,which requires it to fully understand the surrounding environment,be able to deal with complex environmental changes,and quickly locate the target individual in the group of similar objects.Therefore,it is of great significance to build a specific pedestrian tracking system with good real-time performance and robustness.The research work carried out in this paper is as follows:Firstly,a kind of optimization algorithm of YOLO v3 based on mobile net is proposed,and the detection module of service robot tracking system is developed.On the basis of YOLO v3 detection network,the lightweight convolution network called MobileNet is introduced to transform the detection network,which reduces the computation and complexity of the detection model.At the same time,the pedestrian detection dataset is constructed,and the detection speed advantage of the detection model is verified by comparative experiments.Secondly,the overall scheme of the system for the service robot to track specific pedestrians in complex background is formulated,and then the construction of each module of the system is completed.The tracking module of specific pedestrian tracking system is developed based on KCF tracking algorithm.The optimized YOLO v3 detection module is introduced to update and correct the specific target position every 10 frames.At the same time,the target selection strategy is established to locate the occluded pedestrian more accurately,and the occlusion update strategy is established to deal with the complex situation of continuous occlusion.Finally,the experimental verification of the service robot tracking system is completed.The experimental dataset is constructed based on the manually acquired images and open source images,and three different tracking experiments are carried out.In the experiment with pedestrian occlusion,the tracking effect and center pixel error curve show that the tracking system can deal with the occlusion well.In the experiment with pedestrian scale change,the tracking system has the real-time and robustness required by the application scene by evaluating the tracking accuracy and frame rate index.In the actual pedestrian walking video sequence,the tracking results verify that the system also has good tracking effect for the multi pedestrian real scene faced by the service robot.