Probabilistic Lane Association

Abstract: Lane association is the problem of determining in which lane a vehicle is currently driving, which is of interest for automated driving where the vehicle must understand its surroundings. Limited to highway scenarios, a method combining data from different sensors to extract information about the currently associated lane is presented. The suggested method splits the problem in two main parts, lane change identification and road edge detection. The lane change identification mainly uses information from the camera to model the lateral movement on the road and identifies the lane changes as a relative position on the road. This part is implemented with a particle filter. The road edge detection enters radar detections to an iterated Kalman filter and estimates the distances to the road edges. Finally, a combination of the filter outputs makes it possible to compute an absolute position on the road. Comparing the relative and absolute positioning then leads to the desired lane association estimate. The results produced are reliable and encourages to continue approaching this problem in a similar manner, but the current implementation is computationally heavy.