| Under field conditions, water hazards, concave and convex obstacles caused a great threat to Unmanned Ground Vehicles (UGVs), and largely restricted the travel speed. At present, under complex field environment, the three-dimensional lidar is used to recognize UGVs obstacles all day, but the price is expensive and the processing is slow, ineffective. So it is a difficult problem at home and abroad.Therefore, it has very important practical significance for UGVs to identify the ground obstacles fast and effectively.In this paper, based on the current research of UGVs, a new multi-sensor detection method based on line structured light visual sensor system is proposed. It can detection and recognition the obstacles as far as one hundred meters all day. The main research contents and results are as follows:Firstly, the detection and identification system of UGVs for water hazards and the concave and convex obstacles is developed. The hardware system is composed of a line structured light visual sensor system and a position and orientation system. The design of the line structured light sensor is specifically calculated and the principle of detection obstacles is described. Software processing system mainly processes the images gotten by the line structured light visual sensor and LIDAR point cloud data, and fuses data. Through theoretical analysis and experimental verification, the paper has developed the line structured light visual sensor system used for detection and recognition of water hazards and the concave and convex obstacles in the field ground, faraway 100 meters and all day.Secondly, in order to realize the information fusion of multi sensors for obstacle recognition, analysis the characteristics of the sensor and calibrate the sensor system, and set up a complete model of the line structured light visual sensor. On the basis of analyzing the existing line structured light common calibration method, propose a calibration method suitable to the sensor in this paper, give formulas of determine the three-dimensional coordinates of spatial points in three-dimensional camera coordinate, propose and derive formulas of calculating feature size of different obstacles.Thirdly, this paper has studied the image recognition processing of line structured light visual sensor systems, include:the characteristics of the line structured light stripe, the processing of the line structured light strip and obstacle recognition etc. Finally, verify the algorithm by experiments. For the line structured light images, firstly filter the salt and pepper noise of images by RAMF algorithm, then extract light strip region of interest (ROI). Thirdly, on the base of extracting the strip center by Hessian matrix raised by Steger, calculate the parameter in the recursive computation of Gaussian convolution, thereby calculating eigenvalues and eigenvectors of the Hessian matrix. Finally, according to the normal direction of the light strip, calculate the center coordinates. For water hazards, present quick and effective method of "black hole" algorithm. Finally, experiments verify the fast recognition algorithm is effective at different times for different obstacles.Fourthly, fusion the information of line structured light visual sensor and lidar data, because of using a single sensor is difficult to meet the all day and identification obstacles quick and effective in complex environment. Firstly, introduce principles and data presentation formats of the 64-line three-dimensional lidar; with the point cloud library PCL, write the software of the point cloud acquisition and processing by VS2010+Matlab hybrid programming; and the obstacle detection experiments are carried out. After that, identify obstacles based on the D-S evidence theory, and propose the improved synthetic Dempsterrule, and get the D-S fusion output value and the passing dot plot. |
PDF Full Text Request