Detection And Navigation Obstacle Avoidance Methods For Mobile Robot Based On Laser And Depth Camera

With the development of modern robotics technology,People began to consider deploying mobile robots in shopping malls,hospitals,restaurants,and other scenes to reduce the pressure on staff.While performing tasks in these scenarios,the mobile robot is required to reach the destination smoothly while minimizing the impact on the surrounding environment and pedestrians to avoid invading pedestrians' exercise space.This paper focuses on pedestrian detection,pedestrian tracking and mobile robot navigation in multi-pedestrian dynamic environments,and validates it based on ROS robot operating system.The main contents are as follows:(1)The Turtlebot mobile robot platform was remodeled and equipped with an ASUS Xtion depth camera and Hokuyo laser to enable two-dimensional and three-dimensional information of the environment.Then they were calibrated,and the relative positional relationship between the two was determined while correcting the camera distortion.(2)In order to obtain pedestrian information in real time and accurately without GPU acceleration,a joint detection method based on laser and depth camera was proposed.First,the point cloud collected by the depth camera was partitioned,the redundant information was removed and then was projected to the ground to extract the pedestrian area.The pedestrian area was projected back into the depth image,and the local maximum of the outline of the area was extracted,which was used to quickly locate the matching position of the template for pedestrian detection.This method reduces the redundant information of the depth image and improves the detection efficiency.Then use the laser sensor to supplement the depth of view of the depth camera and perform joint detection.By segmenting the acquired laser data,the pedestrian leg features were defined,the Adaboost algorithm was used to classify the detected pedestrians,and finally the Greedy Nearest Neighbor algorithm was used to fuse the detection information of the two sensors,thereby improving the detection range and accuracy.(3)In order to meet the real-time requirements of pedestrian tracking and improve the tracking effect,an Extended Greedy Nearest Neighbor Tracking framework was proposed.For the problem of occlusion,disappearance,and false detection of target trajectories,the Extended Kalman Filter algorithm was used to predict and update pedestrian status.And the pedestrian trajectory management framework is improved,the trajectory maturity and candidate trajectory were proposed.The problem of generation and deletion of pedestrian trajectory was effectively solved,and the robustness to short-term occlusion was good.(4)In order to enable mobile robots to actively avoid pedestrians,Optimal Reciprocal Collision Avoidance(ORCA)that combine with pedestrian motion information was used to local planning.We proposed a method that based on global path can dynamically selecting the ideal speed of robot.The robot establishes an ORCA line for each pedestrian's movement status to obtain a collision-free speed space,and then selects the current optimal speed according to the ideal speed to actively avoid pedestrians and complete the goal.