Study On Recognizing Seat-belt Wear Based On Computer Vision

The number of vehicles is increasing constantly with the rapid development of economy.Meanwhile, the driving safety has been caught more and more attention. As installing driver assistance technologies is an effective measure to guarantee the safety of driving, driver assistance technologies have been noticed and researched by many scholars. Computer vision-based approaches have been regarded as the most promising and valuable techniques among the tookit of driver assistance system. Based on a variety of computer image processing techniques and combined with the geometric features of seat-belt,a new method of recognizing seat-belt wear based on computer vision is proposed. Focused on the problem of seat-belt wear, the proposed approach is realized as a three-step process: image preprocessing and edge detection, fuzzy enhancement, straight-line detection and identification. Furthermore,a large number of experiments are carried out to verify the effectiveness of the method. Main research contents and results of this dissertation are as following:(1) The preprocessing of the collected image includes light compensation and region of interest acquisition. An adaptive revision method based on nonlinear transform is applied to adjust the luminance value of the pixel. Taking advantage of the good inhibitory effect on impulse noise and spot noise, the median filter is used for smooth denoising. Then, the region of interest is acquired. The experimental results demonstrate the effectiveness of this method on balancing the luminance and contrast and reducing the size and noise of the image, as well as the improvement of the algorithm efficiency.(2) After analyzing advantages and disadvantages of several classic edge detection algorithms, Canny operator which could keep the robustness after Illumination changing is selected as the segmentation algorithm to extract edge of the preprocessed image.(3) By analyzing characteristics and disadvantages of the Pal.King fuzzy enhancement algorithm, a new algorithm is proposed based on generalized fuzzy operator GFO which serves to enhance the regional contrast. In order to reduce computations of the Pal.King algorithm, a method to acquire membership is proposed by using upwards semi-trapezoid fuzzy distribution. The experimental results demonstrate that the improved method can avoid losing a lot of gray information and improve the processing speed.(4) Based on the geometric features of seat-belt and combined with improved Hough straight-line detecting, a new detection model of seat-belt wear is proposed. Random Hough transform which has distance constraints is used to detect the straight-line in enhanced image.Firstly,a judgment about the distance condition of the selected dotted pairs is carried out.After acquiring the parameter space, the judgment about the existence of straight-line is carried through. The experimental results imply it has reduced computations and improved the efficiency of the algorithm. Lastly, the judgment combined with conditions like straight slope is carried out to acquire the final result.