Recognition And Distance Measurement Of Obstacles In Front Of Intelligent Vehicle Based On Monocular Vision

Over the years, the field of intelligent vehicles has become a major research theme in intelligent transportation systems since traffic accidents are serious and growing problems all over the world. The goal of an intelligent vehicle is to augment vehicle autonomous driving either entirely or partly for the purposes of safety,comfortability, and saving energy. In the existing context-aware algorithms, machine vision-based context-aware algorithm have an unparalleled advantage because it can provide information on the size and location of the obstacles. Automatic detection of obstacles in front of vehicle has a very important significance to keep the safety distance and prevent the collision accident. The main research contents are divided into four aspects:1.Vehicle road area detection. In the intelligent vehicle vision navigation system,the purpose of the detection of the road is mainly for two aspects, first for estimating the direction and position of the vehicle in the road to control the vehicle according to the predetermined route, second for reducing the algorithm complexity and false recognition rate, and improving the algorithm speed of the subsequent algorithm for the follow-up of the obstacle detection. In order to extract the area where the intelligent vehicles can travel in, an algorithm based on texture feature was proposed,which could be used to distinguish viable path area from the unstructured road. First,in order to extract the texture features corresponding to lane ruts, the Gabor templates of two frequencies and eight directions were selected to analyze the transform of the images, so as to achieve the features of the texture intensity and texture directions of the pixels. And the direction features were used to vote the candidate vanishing points,and the point with the most votes was treated as the final vanishing point. The slopes of lines in the valid vote area was extracted,and the linear equations via the vanishing point were established, so as to get the viable path. Experiments show that the proposed algorithm can divide the viable area effectively in the circumstances ofstrong lights or at nighttime,avoiding the effect of shadows.2.Front obstacle detection. The accuracy of obstacle detection is related to the safety of intelligent vehicle driving, due to the emergence of obstacles are unpredictable and uncertain, so obstacles can only be found in the process of vehicle driving and timely response to the system. An obstacle detection method based on SIFT feature points clustering is proposed, firstly, the ROI region is extracted from the image, and the possible areas with obstacles has been got by detecting horizontal and vertical edges in the ROI region, then, screen and get rid of the regions without obstacles by calculating the entropy of the image of each region, and the SIFT feature points have been extracted in the areas with obstacles, and the position information of the obstacles have been obtained by clustering the feature points using the K algorithm.3.Obstacle tracking based on video image.In the video image, in order to detect the position of the obstacle in real time, the tracking algorithm with Kalman filter and SIFT template matching is proposed to track the position of obstacles. Experimental results show that the proposed method of detecting whether the video object has a target or multiple targets are applicable, and can be implemented by a target to multiple targets and multiple targets into a target transformation.4.Monocular distance measurement. Because of the fast calculation speed and simple structure, monocular vision distance measurement has broad application prospects. The basic theory of monocular vision calibration is introduced in this paper,and finally the exponential function is chosen to complete the fitting on the image of the obstacles in the distance through experimental comparison.