Vehicle Trajectory Reconstruction Using Image Difference Information

Evaluation criteria must be established to guarantee their safety for the widespread use of automated vehicles. One of the evaluation criteria is to reproduce actual near-miss situations and to use the ability to avoid them as an indicator. In this study, we examine the accuracy of image registration used to reproduce near-miss situations. The size of the fixed image is 512x512 pixel. The moving image is a 512x512 pixel area, which is rotated by -60 to 60 degrees in 1 degree increments around the center of the fixed image and translated by 155 pixels in 5 pixel increments up, down, left and right. Image registration was performed for the fixed image and the moving image based on the difference in luminance. As a result, it was found that the images were accurately aligned when the amount of translation was within about 4% of the image size and the amount of rotation was within the range of -30 to 30 degrees.