The Research On Vison Based AGV Navagation Technology With One Single Camera

The vision guided navigation research is one of the most important research areas in AGV (Automated Guided Vehicle) systems. The article mainly aims at the vision applications in AGV systems and discusses detaily the key and related techniques about it. The main topics are listed below:1. Give a brief overview to the development, applications and research status of the computer vision, also give an overview of the development and research status of the vision based AGV systems in and out side of the country. Point out the difficites and new research directions in current vision based AGV research area.2. As a key and basic technique of computer vision, discuss detaily the intric and outtric parametes calibration methods. With the help of the perspective projection model and imaging model, this paper introduces a simple translational motion based intric parameter calibration method using the active motion system. To establish the image points' corresponding relationship in serial images and improve the camera calibration accuracy, this paper introduces a corresponding points' match method based on Harris corner detection algorithm. The experiments prove that this method can calibrate the parameters correctly. Considering the AGV's accural wheeling enviroments and the premise of flat road, this paper introduces inverse prospective projection formula of one single camera which is the theoretical foundation of the image points' space position.3. Discuss detaily the road lane recognition methods based on edge character. After analyze the affects of the sunsine, shadow.etc. to the recognition results, we discover that a better result can be achived using the color feature of the road lane, then introduce an image intensity correction preprocess algorithm based on the color information. Discuss detaily the roade lane edge edge detection methods under the raod lane line model. Considering the image's natural fuzzy behaviour, this paper introduces the fuzzy reasoning based edge detection algorithm. The experiments prove that this method can get the correct results. HT is used to detect the road lane edge, furthermore, this paper predicts the lane's position in the subsequent images according to the current position of road lane in image with the help of vehicle motion model, and a new improved HT is presented which can largely decrease the computation time than HT does. Analyze the colinearity character of the edge points on the line, propose a straight line detection algorithm based on the the smallest eigenvalue analysis, comparing with the HT based line detection method, this method can achieve a good detection result while it is also robust enough to noise and has a obviouse computational velocity increase. This paper also discuss detaily the roade lane edge edge detection methods under the raod lane quadratic curve model. Intoroduce a lane edge detection method based on down projection principle. This method first computes the lane center points' position by alanlyze the road lane's down projection points on the horizontal sample window, then these center points are used to fit a quadratic curve. To improve the detection accuracy of the center points, a false center points detection algorithm based on the lane width is used which has been provd its effectiveness in the experiments.4. Introduce the ordinary steps to detect obstacles on one single image. Under the assumption of consistancy of road gray values, this paper uses FCM (Fuzzy C-means) algorithm to detect the obstacles according the gray value differences between the obstacles and the road. For the FCM's segmentation reslut is sensitive to initialized vlaues, this paper introduces an intensity based FCM's initinal parameters acquisition method. Comparing this with the ordinary cluster validation criteria based method, mehod proposed by this paper can effectively and efficiently detect clusters of various densitites or shapes in a noisy data set of any dimensions. To improve the computational performance of the FCM algorithm, the concepts of the weighting set and weighted sampels are introduced and a novel fast cluster algorithm named weighted fuzzy C-means (WFCM) aogorithm defined on the weighting set is put forward which is used in the gray image segmentation. Theoretical derivation and experiments prove that WFCM can achieve the equivalent segmentation reslut as the FCM does while it has a much better computatioal performance. On the basis of WFCM algorithm, this paper introduces the ordianry fast FCM algorithm: fFCM algorithm. fFCM first uses WFCM algorithm to acquire the adjacent partition result comparing with the FCM's paritition result quickly, then the FCM algorithm is used to generate the final partition result which use the WFCM's partition result as the initial value. Experiments denotes that fFCM can achieve equivalent partition result as the FCM does while decrease the process time at the same time, and it can be widely used in the many field. After the obstacle detection process is finished, the distance calculation formula between the obstacle and the vechile is introduced, and its applicability is also discussed.5. Analyze the motion model of prototype vehicle with three wheels. Design and fabricate this prototype. Use the mature PID controllor to design the vision based motion controllor. Experiments denote that this model and method can satisfy the needs of the vision guided navagation.