Research And Implementation Of Moving Objects Detection Algorithm Based On Fisheye Image

With the rapid growth of the number of cars, traffic accidents are increasing, resulting in appalling casualties and property losses. How to reduce traffic accidents and reduce losses caused by traffic accidents has became a focus of increasing attention. Particularly in backing, due to the blind spots of vehicles rear and the uncertain movement direction of moving objections, it is prone to collision between moving object and ego vehicle. Therefore, give immediately warning before possible collisions between moving objects and ego vehicle, which can effectively reduce traffic accidents. But the moving object detection is affected by the environmental factors, the advantages of feature selection and the accuracy of real-time, moving objects detection is a very interesting and challenging problem.In this paper, we get the video by means of in-vehicle fish-eye camera, due to the image distortion of fish-eye image, its application is still relatively small in the field of of moving objects detection. In this paper, the correction formulas based on planar and cylindrical fisheye images are derived by using the mapping relations between different planes and the transformation relations between different space coordinates. Through the experiments, the results after correction of each algorithm are analyzed, fianally fisheye correction based on the plane is selected.In the stuy of feature point detection and matching algorithms, SUSAN and Harris algorithms are compared in3criterions, stability criterion, antinoise criterion and complexity criterion, Harris selected by comparative analysis. Because Harris algorithm detects feature points which are too concentrated, this paper presents an improved Harris corner detection algorithm. Next, a further study of the pyramid-based Lucas-Kanade of feature point matching algorithm is done, and practical application achieved good results. This paper presents a method of dynamic multi-frame feature point detection and matching based on the study of feature point detection and matching. Through experimental analysis, this method saves running time algorithm and improves the accuracy of feature point matching.In the process of moving objects detection, a further study of fundamental matrix which is algebraic expression of epipolar geometry is done. Based on the study of the traditional fundamental matrix estimation methods, intelligent optimization of the genetic algorithm is applied to fundamental matrix estimation, compared experimental analysis between foundation matrix estimation based on genetic algorithm to previous conventional methods are made, the experiment shows that the genetic algorithm greatly improve the the algorithm stability and antinoise. The motion feature points are detected by the foundation matrix and the principle of epipolar geometry, and then use motion segmentation marker moving objects. Finally, algorithm under different conditions assessment, including different lighting conditions and road conditions, and reach a higher recognition rate.