Research Of3D-camera Based Obstacle Detection In Unstructured Environment For Mobile Robot

Obstacle detection plays an important role in the research of environmentperception field, and is the prerequisite of autonomous navigation. In recent decades,the workspace mobile robots are implemented is expanded from indoor structuredenvironment to outdoor unstructured environment, so, corresponding more betterstandard obstacles testing technology are required. Therefore, based on theunderstanding of outdoor environment having the characteristics of variability,randomness and complexity, this paper studies3D camera based obstacle detectiontechnology in unstructured environment, mainly including three parts:3D cameracalibration and three-dimensional reconstruction, unstructured environment imageprocessing, obstacles features real-time detection.3D camera calibration is the precondition of getting three-dimensional objectinformation. The camera model and the relation between each coordination systemare simply introduced, then the inside parameter matrices of the camera wereobtained through calibriation experiment in this part. Because the unevenness ofground surface, the attitude of mobile robot was changing, we need to use inertialnavigation system to detect the roll and pitch angle of the robot, then the attitude canbe modified. At last, we use3D camera image pixel depth information to reconstruct3d coordinate of corresponding space points. Camera SR-3000vision scope haslimitation. To reduce the use error, exceed7.5m error datas were rectified.On account of the uevenness of unstructured environment image processingparts based on the studies of traditional image processing methods introduces theprocess of image segmentation and interested region extraction taking advantage of3D camera obtainable of gray information and spatial information, then put forward3d information threshold method. At last, edge detection operators were used toextract edge for negative obstacles.Features of obstacles real-time detection parts pointed out the obstacles featuredetection inaccurate problem existing between obstacle and ground region adopting3d information threshold method, and used region recovery algorithm forcompensation. In order to solve the problem of inaccuracy of slope calculationcreated by relative pose uncertainty between mobile robot and slope, slope degreecomputation based on surface orientation and width calculation of ideal ditch algorithm are elaborated, and judged the relative pose between mobile robot andslope through alternative vote method. Then the width computational formula ofnegative obstacles was gived.The last chapter summarizes the achievements and short comings of the paper,the future research work is suggested.