Research On Obstacle Avoidance Strategy Of Binocular Vision For Off-road Robots

Based on the background,The Project Off-Road Robot Contest,the monocular and binocular mixture system is adopted to design corresponding obstacle avoidance strategies for "labyrinth" sections and random obstacle sections and makes the robot complete the task of the road.The main contents of the study are as follows:1.The background and current research status of machine vision technology at home and abroad are expounded.The competition rules of "labyrinth" and random obstacle section of the Project Off-Road Robot Contest are introduced.The solution and key technology of robot obstacle avoidance are expounded in view of the link constraint conditions.2.The parallel binocular system is used,through the “Zhang” calibration method,the binocular camera calibration is completed.In order to reduce the interference of reflected light,it is useful to adjust the light source.3.The vibration of the camera or the heating components would lead to image noise which can impact the feature recognition.Thus through obtaining histogram features by sampling image,the correlation of histogram features can be used to determine what kind of the noise in the image.And finally the ? cut-filter was chosen to remove image noise,compared with the other traditional single algorithm,it has a good effect.4.For the “labyrinth” section,the rough extraction can be completed based on the color and characteristics of shape.Ground can be extracted around the baffles in HSV space,and then the baseboard and baffles in the ground can be gotten.Derived from a monocular camera model can get the linear relationship between the slope of the baseboard and the horizontal distance of the robot.To complete the robot in the section of the lateral positioning,determined by experiments with better accuracy.A binocular system can be used to measure the longitudinal distance of the baffle.The SIFT feature extraction and matching of the crude extraction baffle area to complete the longitudinal distance measurement.The K-means algorithm is used to cluster the baffle features points according to same depth in a baffle to obtain the distance and number of baffles.Through the combination of horizontal information and spatial coordinates of the feature points,the category of the baffle is determined,that is,the left or the right baffle.Finally,experiments show that the strategy can effectively restore the mutual relationship between the baffle and the rack.5.For random obstacles section,the roadblocks are screened by the features of shape to complete the crude extraction.The SIFT feature are extracted and matched in this region to obtain the space location information.Then the K-means clustering algorithm is used to measure the number and depth of roadblocks according to the characteristics of this sections.And then on the built 2D map the artificial potential field method can be used to make a path planning.The experimental results show that the obstacle avoidance strategy can effectively locate the roadblocks and finish the task.