Improvement of the precision of bird’s-eye view record composition with vehicle’s sensor data and camera motion estimation
This study aim at supplying bird’s eye view around a vehicle at parking required high driving skill cheaply. Usual method need to use 4 camera(front, back, left and right) but we only use 2 camera(front and back). In this case, it cannot get left and right side real-time image, so “Bird’s eye Record View(BRV)” fill up this area. BRV is created with the past images which are transformed and composed by sensor data(velocity and angle of wheel) of vehicle. But precision of BRV is worse when this system can’t cope with changing pose of vehicle immediately in wrong way which have a ramp, etc. If driver gets bad view breaking or bending white line of parking space, he causes an accident. So we aim to provide BRV comparable to view of 4 camera system with image transformation which removes the effect of a bad road using camera motion estimate by feature points in sequential images.
© Kohei Saruwatari