Object Detection Based On RGB-D Camera For Middle Size League Soccer Robots

To solve the soccer recognition and obstacle detection in the RoboCup Middle Size League competition,the algorithm based on RGB-D camera is designed.The precise three dimensional localization information of the soccer is provided by the vision system in real-time,which lays the foundation for successfully playing football by the robot.The perception ability of the Middle Size League soccer robots can also be improved,and the dependence of the soccer robot on the color recognition is reduced.In this paper,the detection algorithm in robot soccer is proposed based on RGBD camera.The first step is save the RGB image by Kinect,then calibration color and setup a color lookup table.According to the calibration result,the soccer candidate area is extracted in the point cloud data.The second step is to fit the three-dimensional sphere model by RANSAC algorithm,and the soccer region is segmented to complete the soccer detection in three-dimensional space.Then,an algorithm of obstacle detection is proposed based on 3D point cloud.The point cloud data obtained by Kinect is adopted to remove the background information by threshold segmentation in order to obtain the foreground of including the target.Then,the point cloud data are reduced by downsampling which is benefit in increasing the detection speed.After that,the plane model is fitted by the RANSAC algorithm,the objects are extracted out of the point cloud data by the nearest neighbor clustering algorithm.Lastly,the objects' geometric information are used to re-segment the objects.Overall,this algorithm improves the object detection accuracy and can detect the obstacle efficiently.Finally,based on the soccer robot omnidirectional vision system and RGB-D camera,an adaptive weighted fusion algorithm is designed.In the case of minimum mean squared error,the data fusion value of the current sensor is obtained by using the estimated value of the previous time and the current measured value.The result shows the method improves the accuracy of the visual system,and provides accurate location information for velocity estimation,robot motion planning and control,etc.