Research On All-Around-View Parking Assistant System Based On DSP Processor

With the rapid development of national economy, road congestion makes it increasingly difficult for drivers to park due to the considerable increase of vehicle ownership. Blind area of the vehicle proves to be the greatest obstacle to parking. According to the statistics,30% of the accidents are caused due to the rear blind area. This article studies and designs an all-around-view parking assistant system based on DSPs. The proposed system captures real-time video through cameras mounted around the vehicle, and then outputs the vehicle’s 360 degree panoramic bird view image processed by embedded DSP platform. The driver could park safely and effectively by observing comprehensive information around the vehicle without blind area, which has positive significance for improving security of parking.This thesis firstly describes the system architecture and expounds the rationality of the 4 cameras’installation location. Then configuration of the multi-level memory architecture and video codecs are made by taking full advantage of the hardware resources integrated on DM642. For real-time multi-channel video capture and display, this article designs the hardware driver and upper application program based on DSP/BIOS operating system, which construct the all-around-view parking system successfully.On analyzing the imaging principle of the camera, this thesis establishes a fish-eye camera imaging model and presents the transform matrix between the image and the scene. This paper uses radial symmetric model which is different from the conventional pinhole model. Since the fish-eye lens would introduce serious distortion, this paper studies the types of fisheye image distortion and proposes an improved correction model, which is on the basis of the division model. As a necessary prerequisite for calibration of the camera parameters, this paper studies the Zhang Zhengyou calibration algorithm and general calibration algorithm. Combined with the characteristics of fish-eye camera, the general calibration algorithm is selected to calibrate the camera, which could achieve high precision.On the basis of capturing and correcting the multi-channel video, this thesis generates a top view of surroundings which conforms to the designed interface through spatially transforming the multiple images around vehicle to unified virtual bird view coordinates. As a key step in constructing the panorama effect, this thesis analyzes the multi-source image stitching technology. Considering the resource constraints on DM642, the multi-source image around the vehicle is manually corrected. The overlap region is blended and brightness balanced according to the actual situation. Experiments show that the registration and blending method performs well.This thesis implements an all-around-view parking assistant system based on DM642, and optimizes the hardware and software. Synthesizing the key technologies and algorithms proposed above, this paper implements the algorithm optimization and performance improvement by looking-up table and dual-channel buffer scheduling and introduces the C6000 platform programming technology. Experiments show that all-around-view parking assistant system designed and implemented in this thesis provides a clear real-time surrounding scene for drivers in bird view without any blind area, which provides effective assistant for drivers to achieve secure parking in narrow and crowded roads.