Automatic image deviation detection for AVM auto-calibration

Abstract

Around View Monitoring(AVM) images are widely used in Advanced Driver Assistance Systems(ADAS). In order to generate an AVM image, we need to obtain four coordinates from the front, left, right, and rear camera images installed to a vehicle to perform perspective transform. Two coordinates can be extracted from the car body and the other two can be extracted from lane end points of the images. However, due to external factors such as collision, drift, etc., the physical position or angle of the installed cameras may change. This leads to the corruption of the initially obtained coordinate's actual location inside the image. In this paper, we propose an AVM auto-calibration algorithm which uses automatic image deviation detection. We compare the current car body coordinates to the initially obtained car body coordinates for deviation detection. For four 640×480 input images, it takes 2400ms for deviation detection and 3875ms for the whole AVM auto-calibration algorithm at Intel i5 3.5GHz processor environment.