Obstacle Avoidance Strategy Of Nurse Assistant Robot Based On Pedestrian Tracking And Prediction

With the development of computer technology,artificial intelligence and sensor technology in recent years,a series of mobile robots such as driverless,sweeping robots and feeding robots come into being,and mobile robots begin to slowly enter our life.Due to the particularity of the hospital environment,the current work status of medical staff is under the intense work pressure.If the mobile robot is applied to the hospital environment to help the medical staff to carry out the handling work within their power,the doctors and nurses can release from the heavy and simple handling work.Therefore,this paper studies the methods and strategies of dynamic obstacle avoidance in indoor environment.Based on the traditional obstacle avoidance methods,the following obstacle avoidance processes are proposed: pedestrian recognition and tracking based on depth image,and after the avatar angle judgment and path prediction,preliminary judgment Pedestrian movement trajectory,and then put the pedestrian movement trajectory into the path planning of mobile robots in the future,so as to better achieve the obstacle avoidance behavior.Specific work as follows:Pedestrians are identified based on RGBD images captured by Kinect V2 sensors.Extract possible areas for pedestrians and perform body recognition in the same area of the color image.Based on the HOG feature and SVM classifier to identify the environment of pedestrians,improve the detection accuracy and computational efficiency.Then the acquired color image is down-sampled,pedestrians are identified and time-stamped on each layer of the image.Then match the next identified sample with existing pedestrian model orb feature,high degree of matching is considered the same pedestrian.Then the B-spline fitting is performed for discrete position points of the same pedestrian to obtain the movement trajectory of the pedestrian in the past period.Recognize of pedestrian head angle based on one-to-many SVM classifier.Because of the application in this paper,the prediction of pedestrians is between 3 m and 5 m,the image feature extraction is poor,so the traditional face detection algorithm does not apply.In this paper,based on the SVM classifier in machine learning,Haar features are extracted and classified according to the human avatar angle.Then the avatar angle classification of the detected pedestrian frame is performed to complete the pedestrian motion prediction.POMDP Framework Based on Deep Learning Behavioral to plan for Robot's Speed.The original moving robot's speed is based on the planned path points for speed planning,and does not consider the environmental impact outside the path point.Therefore,when the pedestrian suddenly appears in front of the robot from the side,a collision caused by the avoidance can occur.In this regard,this paper describes the environmental abstraction as Markov random field,so that when the pedestrians appear in the planning area,they can accelerate or decelerate in advance to solve the problem of evasion in time to better complete the robot's dynamic barrier.Experiments show that the dynamic obstacle avoidance path planning method designed in this paper is more stable and intelligent in sparse population and safer in the process of mission.