Research On Unmanned Vehicle Tracking System Based On Target Vehicle Trajectory Prediction

The rapid development of technologies such as big data,artificial intelligence,and mobile computing has greatly promoted the improvement and development of target tracking technology.Target tracking technology mainly relies on sensors such as vision and radar to sense the motion state of the target.However,the type of target information obtained through these methods is relatively single.After the target is occluded by other objects and out of the sensing range,it is difficult to retrieve the tracking target after it is lost relying solely on sensors such as vision and radar.Faced with complex mission requirements,unmanned vehicles have the advantages of high task execution efficiency and strong reliability.Aiming at the problem that targets in complex scenarios are difficult to be retrieved after they leave the sensor’s sensing range,this paper proposes a new assisted tracking strategy,which predicts the trajectory of target vehicles based on GPS positioning technology,and schedules unmanned vehicles to assist in tracking the predicted area,in order to more effectively and reliably complete the target tracking task.The main research work of this article is as follows:(1)A target vehicle trajectory prediction method based on spatiotemporal attention(TPred)is proposed.This method is a vehicle trajectory prediction model based on Transformer and graph attention network.The specific prediction process can be divided into trajectory preprocessing module,spatiotemporal feature extraction module,and future trajectory prediction module.Firstly,a map matching algorithm is used to convert GPS track records into road segment sequences.Next,TPred converts the road segments in the road network into local directed graphs based on the driving direction,extracts spatial features in the road network through the graph attention network,and uses Transformer to extract temporal features in the input sequence,improving the efficiency and prediction accuracy of the model.Then,a Filter layer is introduced to filter the model output using a local directed graph to output a continuous future trajectory sequence.Finally,this paper verifies the prediction accuracy of the model on two data sets in Beijing and Chengdu.The AMR value of TPred on the Beijing dataset reached 73.07%,which is 37.45% and 6.38% higher than that of LSTM Encoder Encoder and Transformer,respectively.On the Chengdu dataset,the AMR value of TPred reached 78.93%,which was 36.06% and 10.55% higher than that of LSTM Encoder Decoder and Transformer,respectively.(2)A vehicle scheduling method based on graph matching is proposed.The method is divided into two parts: a vehicle scheduling matching weight determination method based on multiple indicators and a two-stage unmanned vehicle dynamic scheduling method,which solve the problem of difficult to determine the weight of multiple indicators and the dynamic nature of vehicles and tasks in the vehicle scheduling process.Firstly,the entropy weight method is used to calculate the weights of multiple indicators to determine the matching weights between different vehicles and different tasks.Then,a two-stage vehicle scheduling method is used to match the unmanned vehicle and the task.When a task first appears,a greedy algorithm is used to assign the task to the available unmanned vehicle with the largest weight.Subsequently,a time window based rescheduling method is used to rematch all tasks to be executed with reassignable unmanned vehicles to fully utilize the resources of unmanned vehicles and achieve global optimization.In order to verify the effectiveness of the scheduling method,this paper designs and builds a tracking control platform,and conducts experimental tests on the tracking control platform.Experimental results show that the dynamic vehicle scheduling method based on graph matching improves the performance and efficiency of the system.(3)A target tracking system based on unmanned vehicle assisted tracking is designed and implemented.This system consists of two parts: a unmanned vehicle tracking and control platform based on Baidu Maps and a unmanned vehicle hardware platform based on ROS.During the process of tracking the target vehicle,the tracking control platform predicts the future trajectory of the target vehicle by tracking the location records reported by the vehicle,and dispatches unmanned vehicles to the prediction area to assist in searching for the target vehicle.In order to verify the feasibility of the method proposed in this article,we used unmanned vehicles to simulate target tracking scenarios,and conducted experimental tests.Among them,the AMR value of TPred on the unmanned vehicle trajectory dataset reached 78.52%,which was 21.69% and 13.61% higher than that of LSTM Encoder Decoder and Transformer,respectively.