Design And Implementation Of Monocular Vision-based Obstacle Detection And Path Planning For Sweeping Robot

Sweeping robot is currently playing a dominant role in household service robots.Compared with other robots,sweeping robot is well industrialized and applied.The pivotal technologies of sweeping robot include obstacle detection,path planning,automatic recharge,etc.Among these features,path planning is a key factor to evaluate the intelligence level of the robot,which means all areas required for sweeping should be covered.For this reason,obstacle detection and path planning has been studied in this thesis.The premise of path planning is to detect the environment,so obstacle-detecting sensors are analyzed at first.Then a ranging scheme of monocular vision sensor combined with line structured light is determined.In addition,Histogram-based threshold segmentation and the method of grayscale features in cross-section are designed to process line-structured light images in indoor environment.The measurement formula is given after calibration.The experiment shows that the average error of measurement is less than 1cm within 50cm,and the measurement frequency is about 10Hz.In this thesis,a complete coverage path planning algorithm based on cellular decomposition is designed to ensure the sweeping robot to traverse the sweeping area completely.The central essence of the algorithm is area sweeping-missing judgment-area connection-area sweeping.In area sweeping,common "zigzag" sweeping rules is improved to make it more adaptable to actual environment.A state machine based obstacle-edge-follow algorithm is designed for obstacle avoidance.The algorithm summarizes the obstacle-edge-follow process into three states.Besides,linear and polygonal models are used to calculate the information,for example,relative angles between robots and obstacles.Then there comes the state transition conditions.The improvement of the sweeping rules and the design of obstacle-edge-follow algorithm enhance the sweeping of obstacle edges.Once an area is cleaned,missed area is determined according to environmental map.The missed areas are classified and the appropriate starting point for each missed area is selected.D*algorithm is applied as the basic algorithm for the point to point planning of area connection.The algorithm plans a path offline according to the environmental map and can perform online re-planning when the environment changes.Considering the robot's turning consumption and the safety of the path,path cost function of modify D*algorithm is modified with the use of distance transformation.The designed algorithm is implemented on the embedded robot platform,and many sets of experiments in different environment are conducted.Experimental results show that the response time of obstacle avoidance is less than lms.Average online re-planning time of point to point planning is about 200ms,and coverage percent is no less than 93,which proves the effectiveness of the path planning algorithm in this thesis.