Research On Traversable Region Detection Based On Hybrid Vision System For Mobile Robots In Outdoor Environments

For mobile robots in complex outdoor environments,recognizing the traversable region is the key problem of route planning.It needs to recognize the traversable region not only in front of the robot,but also in every direction around it.Combining the parallax information of binocular vision system and the panoramic image information of omnidirectional vision system,this paper proposes binocular-omnidirectional hybrid vision system to recognize traversable region in 360 ° environments around the robot,which lay a foundation for robots' autonomous navigation in outdoor environments.Firstly,in order to recognize the traversable region in front of the robot autonomously,we choose the Bumblebee binocular sensor.For calibrating the parameters of omnidirectional vision system easily,we choose hyperbolic mirror and fix it with a Prosilica industrial camera to be a omnidirectional vision system.Aiming to get a large enough common field of view,we fix the binocular vision system in front of the omnidirectional vision system in an omnidirectional mobile robot.Then,with the Elas matching algorithm,the parallax of the binocular vision system can be computed,and the traversable region in front of the robot can be recognized by U-V parallax algorithm.For mapping the result to panoramic image,this paper proposes the image mapping method based on coordinate system conversion.By a world coordinate system,the mapping relationship between the left image coordinate system and the panoramic image coordinate systems can be computed.The parameters of the coordinate conversion can be computed by MATLAB calibration toolbox.And then,SVM algorithm is chosen for machine learning with the sample of feature and label of traversable region and intraverble region if front of the robot.In order to improve the efficiency of the algorithm,the panoramic image is segmented by Preemptive SLIC algorithm.Then LBP feature and WLD feature are chosen for classification,so that the traversable region in 360°environments can be recognized.Finally,a series of experiments in outdoor environments are performed.The results show that the algorithm can expand the range of traversable region dection,and more perception information can be provided for robots' route planning.