Transponder- and Camera-based advanced driver assistance system

Abstract

Cooperative traffic safety is a straightforward approach for a significant reduction of accidents and fatalities. This paper presents a predictive safety system based on a cooperative localization technology using transponders combined with a monocular camera. By means of these sensor components other traffic partners in the surrounding area are recognized and tracked even in case of occlusion. Using the pedestrian detections of the transponder system for the generation of regions of interest (ROI), video-based confirmation is achieved in real-time using histograms of oriented gradients (HOG). An extended Kalman filter is applied to cope with adapted nonlinear process and measurement models for transponder-based tracking, including methods for compensation of the vehicle's ego motion and sensor mounting offsets. The collision risk with other traffic partners especially pedestrians is assessed by using sophisticated motion models based on empirical data. In an experimental study of real-world scenarios it is demonstrated that the fusion of the sensor data results in a reliable prediction of upcoming collision risks and enables a specific warning or a justified autonomous brake maneuver in order to avoid a collision. The results confirm excellent detection, tracking and real-time performance and emphasize the potential of transponder-based active safety systems.