Method Research Of Vehicle Detection And Distance Measurement Based On Machine Vision

With the rapid increasing of car quantity, traffic safety problem has become more and more serious. Safe interval warning technology based on machine vision has some advantages, such as small visual system, easy setup, low cost, and popular foundation technology platform for image processing and so on. Nowadays, it has become a research hotspot in intelligent transportation system. In this thesis, the vehicle on the highway is the research object, which is detected by improving K-Means clustering algorithm, recognized with improved Hu invariant moment and improved Affine invariant moment, and then tracked using Camshift algorithm. Then, the method of data fitting is applied to calculate the distance based on the vehicle detection and tracking. The main research contents are as follows:1. The methods of wavelet threshold denoising and median filtering are combined for image preprocessing after theory analysis and contrast tests because most of images contain Gaussian noise and salt pepper noise. Based on detecting the driveway lines with Hough transform, K-Means clustering algorithm is modified at the aspects of clustering center, clustering number and distance measure aiming to detect the vehicle accurately and avoid the defects of instable results and divergence of classic K-Means clustering algorithm.2. Moment character has the invariance property when the image is rotated, translated, scaled, and is used in the field of image recognition widely. Hu moment has no scaling invariance property under the circumstance of discrete data, so it has to be improved. Moreover, improved Hu moment and Affine moment have to be processed with logarithm in order to limit big changing range of the data. Rectangle degree, symmetry, improved Hu moment, improved Affine moment characters of the image are utilized synthetically. The image is processed according to the thresholds of rectangular degree and symmetry, and then matches with templates of combined invariant moments of improved Hu moment and Affine moment to identify itself.3. In order to measure the distance fast and effectively, the method of data fitting is calibrated to calculate the distance based on the foundation of vehicle position which is tracked by Camshift algorithm, and the method of presupposed safe interval warning is applied to reach the aim of giving drivers different signals based on the different distance range.4. To test the effectiveness of above proposed methods for vehicle detection and distance measurement, the thesis processed the real car experiments for different environment and different vehicle types based on the actual highway conditions. The experimental results show that the proposed methods for highway vehicle detection and distance measurement are feasible.