Research On Travel Planning Based On Individual Movement Prediction Of Crowd

Aiming at the scenarios where robots need to travel among pedestrians,this thesis determines the main technical route based on detection-prediction-obstacle avoidance,and improves its main algorithm to improve its calculation speed,prediction accuracy,and obstacle avoidance effect,so that the whole system can better complete the dynamic obstacle avoidance task for pedestrians.In the detection process,this thesis uses the point cloud as the perceptual data,and uses the algorithm flow of point cloud preprocessing-point cloud semantic segmentation-point cloud instance segmentation-3D target regression-3D target tracking to complete the detection of pedestrian position,posture and speed.In the point cloud preprocessing,a probabilistic screening sampling method based on point cloud density is proposed to improve the efficiency of point cloud processing in the lidar environment.In the point cloud semantic segmentation,a point cloud semantic segmentation method based on block feature fusion is proposed,and the experimental results of the published dataset show that the point cloud segmentation speed is improved by 2 to 4 times while ensuring the accuracy of the segmentation.In the prediction stage,this thesis uses the improved social force model as the basis,the input of which is the position and speed of the pedestrian obtained in the detection link,and the parameters in the social force model are obtained by the multivariate nonlinear optimization fitting public data set to predict the average speed of the pedestrian over time.Comparing the optimized social force model with the time series prediction model based on deep learning,the results show that the prediction error of the social force model is 23.4% higher than that of the time series prediction method,but the generalization performance of the social force model in different data sets is strong,which can deal with the short-term prediction problem of pedestrians.In the obstacle avoidance section,the VO-based obstacle avoidance algorithm is used as input,and the pedestrian velocity obtained in the prediction link is used as input,and compared with the classical dynamic window method on the public data set,the results show that the success rate of obstacle avoidance based on VO is high.At the same time,compared with the VO algorithm based on linear pedestrian prediction results,the accuracy and effectiveness of the pedestrian prediction model in this thesis are proved.