Research On Pavement Survey Based On Omnidirectional Vision Sensor

OVS(omnidirectional vision system, also called omnidirectional camera),whichhas a horizontal viewing angle of360°for imaging in a single shot, has been widelyused in monitoring、SLAM(simultaneous localization and mapping)、mobile robotand3D digital city. One kind of such camera, called catadioptric omnidirectionalcamera (short for COC) featured simple structure、 low cost and convenientmaintenance, is becoming main orientation of OVS development. Traditional visionlocation system using ordinary camera, which has narrow sight, is not beneficial forrapid imaging large-scale scene. In order to optimize system structure, improvesystem-processing speed, OVS is put forward, it has simple structure、convenient to fixand used in many fields.The paper introduced technology of panoramic imaging, analyzed features ofseveral kinds of panoramic camera, compared different catadioptric omnidirectionalcameras, the development of panoramic vision location is summarized, the system forpavement survey basing on panoramic vision location is proposed.The theory of catadioptric omnidirectional camera was derived, several kinds ofmirror such as conical surface,spherical surface, hyperboloid surface, ellipsoidalsurface and parabolic surface was simulated, the results of simulation showed that thecatadioptric omnidirectional camera consisted of constant horizontal resolution mirroror concave hyperboloid mirror is suitable for pavement survey. Compared differentstent for COC, designed and put up a COC, which is960×960pixel resolution、30meters diameter for horizontal scene、constant horizontal resolution and optical axisadjustable.Pavement was segmented from image basing on region growing algorithm, thepavement’s direction was calculated by expansion, edge detection, Hough transform andsome other image processing technology, finally, the course angle was indirectlyachieved. Through static test, reference for measurement results of laser theodolite, for10-sample size and at0.25significant level, the new method’s measuring deviation issmaller than0.35degree.Monocular visual odometry system was put up basing on COC. Reference fortraditional method of monocular visual odometry (extracting feature points by SURFalgorithm, matching feature points by exhaustive algorithm, estimating similartransform model by RANSAC algorithm), trajectory formula was derived. Analyzedand indicated that accumulated error of matching images is the main reason for limitedaccuracy of monocular visual odometry, the method for correct course angle in straightpavement was putted forward, the trajectory of the moving platform measuring experimental shows that, the trajectory is more accuracy after correction. Compared totraditional method of monocular visual odometry, new method has1.7meters lateraldeviation,1.8meters vertical deviation in150meters’ driving distance, in order toimprove real-time performance, course angle correction combining images matchingalgorithm was put forward, and positioning speed acquire great improvement.