Car-like Mobile Robot Path Tracking And Obstacle Detecting System Based On Monocular Vision

Wheeled car-like robot is an important branch of mobile robot. Its navigation problem isalways hot topics in science research. Vision is the most primary source to acquire theinformation of surroundings, and it can also provide information of environment to robot.This paper discusses path tracking and obstacle detection which are two basic problem ofnavigation based on monocular vision image processing. The system’s input is the imagesequence grabbed by the camera on the robot in motion. Comprehensive uses of the distance,angle and path winding level to realize path tracking. At the same time, locates obstacle byoptical flow distribution and three-dimensional information.The entire system is divided into three parts: vision sensor part, image understanding anddecision-making part, motor execution part. The vision sensor part is used to grab the imageof robot’s forward environment. Image understanding and decision-making part is mainlyresponsible for three tasks. The first is image preprocessing. Second is acquiring useful sceneinformation by specific image processing and recognition algorithm. The last is outputintelligent decision command. Motor execution part has the function of control steering andspeed of robot after command analysis.The majority of research includes:(1) Introduce basic preprocessing method include image gray processing, filtering andhistogram equalization. These operations ensure the algorithm’s robustness even in specialenvironment such as light or dark and provide quality image for path detection and opticalflow calculation.(2) Path detection and tracking is the main point in this dissertation. Add speed fuzzycontroller on the basis of angle controller from humanoid driving perspective to slow downwhen the path has large curvature or the robot’s present heading angle has large differencefrom path direction. This method is applicative to both straight and curve path. According tothis design, propose the way of present path by start and pre-view points, which start point’scoordinate range is calculated through the incorporation of threshold segmentation and edgedivision, pre-view point is determined by means of edge tracing.(3) Propose a method to separate the region of obstacle based on sparse optical flow.Compared with dense optical flow, the sparse one is more suitable for robot because thecalculation must be efficient as well as accuracy. As the difference of depth, the average valueof optical flow at obstacle is larger than surrounding. Extract optical flow information ofcorner points through pyramid Lucas-Kanade method. And the possible position of theobstacle was estimated with improved balance strategy. The obstacles with the width not cover the camera view can be detected correctly in static scene.(4) Calculate focus of expansion and time to contact form optical flow, then sketchyestimate relative depth. This can be used to assist in obstacle detection.(5) Build simulation and real experiment platform. Display the effect of robot control andmove entirely and visually. At the same time prove the feasibility of the algorithm.