Vehicle Safety Assistant System Based On Panoramic Vision

With the growth of China’s economy, the total number of cars is increasing rapidly.Roads and packing lots are getting more and more congested. Drivers often have to drivethrough narrow streets or park in narrow parking lots, in which drivers’ field of vision islimited and minor accidents are likely to occur, causing unnecessary damage. Among all kindsof vehicle safety assistant systems, the system based on panoramic vision can provide thedriver with real time aerial view video, which eliminates all the blind spots around the vehicle.This dissertation studies the key technologies of panoramic vision based driver assistancesystems, and designed an implementation of the system: install up to four fish-eye cameraaround the car body, then run the fisheye image distortion correction process; transform theperspective view image to top view image; and finally run the image mosaic algorism togenerate the objective aerial view image. This dissertation presents the design of our vehiclesafety assistant system based on panoramic vision, including the following achievements:(1) According to the relationship between fish-eye camera model and pinhole cameramodel, we designed a fish-eye image correction method, which transforms distorted imagescaptured by the fisheye camera into a perspective image of the pinhole camera.(2) We used three-dimensional coordinate system transformation method to convert theimage perspective, and proposed a parameter correcting method. In this way, we transform theperspective images into top view images.(3) We generate the mosaic image according to the relationship between cameracoordinate system and the world coordinate system, and proposed and image mosaicalgorithm based on the matching of vertical distance of edges, which achieved automaticimage stitching.(4) According to the relationship between the spatial relationships between the imagesfrom different directions, using9-point interpolation, we established the pixel address lookuptable. When run in real time, the system completes all the transformation steps by looking upthe lookup table, which allows the system to provide real time video.By studying the existing relevant image processing technology, we has made some improvements based on the prior art. We introduced an automatic image mosaic algorithm andthe9-point image interpolation method. The system we designed performs well in terms ofinstallation adjustment, image quality and video fluency. Real vehicle experiments areconducted to prove the effectiveness and timeliness of the system.