A Joint Integrated Probabilistic Data Association Filter for pedestrian tracking across blind regions using monocular camera and radar

Pedestrian tracking in advanced driver assistance systems in commercial vehicles is not only important in the frontal field of perception, but also in the blind spot of the vehicle (right side), e.g., to mitigate or avoid collisions during turning maneuvers. While a camera system and radars observe the front, only a radar is available at the vehicle's side. This paper will present a Joint Integrated Probabilistic Data Association Filter (JIPDAF) that tracks pedestrians in the frontal field of view and in the vehicle's blind spot. Although the sensors do not have a common field of view, we show that tracking across the blind region is advantageous, since information that has already been retrieved by the front sensors can be conserved, and the confirmation time of the tracks could be reduced. The results include a comparison of the JIPDAF approach running in real-time with an extended Kalman Filter with global nearest neighbor data association using data from real measurements. Furthermore, we will compare the fusion results to measurements of a 3D laser scanner. To the authors' knowledge, there is no JIPDAF approach for pedestrian tracking using camera HOG detections and radar sensors yet.