Pavement Target Recognition And Application Based On Radar And Image Information Fusion

Unmanned Ground Vehicle(UGV)has important research value and a wide range of applications both in the military and civilian areas.Perception for typical road environment is one of the key technologies of UGV.Multi-sensors can achieve more accurate and efficient perceptive effects due to their complementary perfonnance.The aim of this paper is to fuse the multi-line radar data with the image data,and to explore the two sensors to identify the targets on the road by UGV.The main contributions are given as follows:Based on the 32-line radar,the raster map is constructed by the relative height difference of measured data.Then a density clustering algorithm based on the distance pre-judgment is proposed.The areas where have intensive points are the obstacle areas.Non-pavement barriers can be filtered through prior knowledge of geometry.The experimental results show that this method can successfully detect the road obstacle and get the distance of obstacle.This paper also studies the pedestrian detection based on image.Firstly,the Gaussian model is used to model the background to filter the most of the background.Secondly,the erosion and dilation of the morphological filter algorithms are adopted to remove the isolated dots and fuse the adjacent dots.Finally,the rectangle contour is used to find the obstacle position.The obstacle detection algorithms based on image and radar are compared.At the same time,the advantages and disadvantages of these two sensors are analyzed.Multi-sensor information fusion is based on the unified data coordinates.First,according to the corresponding corners of the calibration objects,the parameters of the radar are calibrated in order to find the conversion matrix of radar data points from the original polar coordinate to the world coordinate.Then the camera is calibrated by Tsai method,and finally the projection matrix of the radar coordinate system to the image pixel coordinate system is obtained.A method of data fusion based on obstacle is proposed.The depth of obstacle detected in the radar is fused with the image data so that the pixel of the image contains not only RGB information but also distance information,which is ready to identify the target and determine whether the target distance is safe.It is of great significance for the safe driving of UGV to confirm the position of obstacle on road,especially the position of pedestrians.The HOG feature operator is extracted from the region of interest on the image,which is extracted by radar detection at first.The pedestrian data is taken as the positive sample,and the real driving scene of the UGV where pedestrians are not included is taken as the negative sample.Then the positive samples and negative samples are trained for linear SVM classifier to identify whether the obstacle is pedestrian.Experiments show that the method can realize the real-time target recognition and can be applied to the autonomous navigation system of the unmanned vehicle,which has a great significance of practical application.