Panorama Parking Assist System Based On SIFT Algorithm

With the rapid development of Chinese economy, the number of car is also significantly improved, while traffic safety problems are more prominent. To solve this problem, some driver assistance systems are applied on the vehicle safe driving.Parking assist system has panoramic visual information around the car, presenting information around the body on the display device in the car in real time. This system eliminate the visual blind spot, the driver can keep abreast of the road conditions based on the system, then the driver can make the appropriate adjustments, especially when driving in the parking and narrow path, this system is even more important. So scholars and automobile manufacturers are carried out research in this field, and get great results.In this paper, on the base of deep research on the current situation of domestic and abroad, found that the relevant research and most of the parking assist system on the market still exists seam, and has poor visual effects, so this paper ultimately achieve seamless Stitching as the target of the system to improve the visual effect. Firstly, this paper gives an overview of the domestic and foreign research situation and development trend, analysis of the technical points of panorama parking assist system, the key technology of the system elements as lens selection and it’s installation position and angle, the fisheye image distortion correction, image plan form changes, seamless image stitching, and the solutions of the technology are given in this paper. The fisheye camera is selected as the image acquisition equipment, and installed in the four directions around the car body, the front and rear cameras fixed with an angle of 30 degrees between optical axis and the vertical, and left and right cameras fixed with the optical axis perpendicular to the floor, so the camera can be obtained directly topview around renderings, reduce the time consumption of overlooking transformation; For fisheye lens distortion correction, the paper analyzes the advantages and disadvantages of several typical correction method, a distortion correction method based on double longitude spherical model fisheye distortion corrected method is proposed, and obtain better results; image stitching, this paper select the image registration method based on SIFT, by alignment of the four pairs of images, and real-time of traditional algorithm is improved, greatly reduce the time consumed, which can meet the needs of practical application, and finally through the seamless fusion gradually fade out weighed fusion algorithm to realize image, get good effect.Research contents and achievements of this thesis mainly include the following aspects:1) Based on the traditional fish-eye image distortion correction method, a distortion correction algorithm based on double longitude is presented:First of all, when the camera with unknown fisheye angle and image size, obtain the spherical center and radius by using the radius of fisheye image, and Optimize radius to weaken pole distortion; on basis of this,then using the double longitude model method, the fisheye image is mapped onto the sphere by orthogonal projection strategy, transverse and longitudinal into spherical longitude coordinates, thus transverse and longitudinal direction double longitude coordinates projected to the square planar image, rapid, accurate image transform. Finally, to verify the effectiveness of this proposed algorithm and the feasibility of using two examples;2) The traditional SIFT-based image registration algorithm was improved:feature points matching methods such as value through bilinear interpolation scaling, feature point extraction in overlap region and the extreme characteristics of single point registration and other methods to achieve fast image registration; then zooming back and forth according to the image Contact between the transformation matrix to obtain registration between the original large-size images, proven that the algorithm effectively improves the registration rate.3) Using the fading in and out of the image registration algorithm after the image seamless fusion of registration:after four image overlap region pixels, using weighted average gradually fade out method, the weight value overlap region boundary is directly related to the distance in pixels, obtain better fusion effect.