Motion estimation for 1-Channel around view monitoring in advance driver assistance system using wide fisheye camera

Abstract

Around View Monitoring (AVM) system uses multiple input cameras mounted on different vehicle positions to display 360° bird-eye-view around the vehicle that is not readily visible to the driver. The development of this system will reduce parking accidents by monitoring its surroundings, detecting lanes, and identifying obstacles. Even a short propel guidance can diminish the number of accidents, even minor ones. This project proposes developing one ultra-wide-angle camera on the rear vehicle integrated with the motion estimation (ME) algorithm to produce a parking bird-eye view. This algorithm will not depend on other sensors such as GPS, odometer, and steering. The AVM system must be fast enough to make the image as a bird's eye view and make it as close as possible compared to the real world. The algorithm will use information from ME to stitch image sequences captured from the front or rear of vehicles. Hence, it will create a synthetic image around the vehicle for the AVM system. With this solution, all kinds of vehicles will have AVM technology, even the old vehicles. Before applying ME in the AVM algorithm, the images will undergo pre-processing, which are dewarping, top-down view, and cropped. ME is needed to calculate vectors that show the motion of the vehicle. The studies show that the ME method that can be used for a homogeneous surface is indirect. From the indirect method, there are optical flow and block matching. After the analysis, optical flow is deemed unsuitable for a real-time ADAS system as it fails at least 25.5% of the time. This statistic means, out of 100 frames, the algorithm will fail at least for 25 frames. Usually, the real-time application is up to 30 frames per second. 25 frames are close to one second, and an error of one second is unacceptable. On the other hand, for block matching, the results for percentage fails are 6.76% and 18.24% for left and right segments of the images. The ME methods result are analyzed, and the block matching method fits the system with the highest accuracy and lowest processing time. Thus, the new algorithm for the AVM system is based on the Block Matching method. This project’s algorithm is on par with CCORR_NORMED with 14.86% and 18.98% fail percentages for left and right segments, respectively. CCORR_NORMED processing speed is 4.323ms, while this project’s ME is 4.655ms. PC platform produces 12.750 ms processing speed with 47 fps, 94.594 ms with 10 fps on Renesas, and 172.955 ms with 5 fps on Telechips.