Institut für Mess- und Regelungstechnik (MRT)

Pedestrian intention recognition for active safety systems

Motivation

While recently the number of people injured in traffic accidents has been decreasing, the number of injured pedestrians remains nearly constant. In order to reduce the number of injuries, active safety systems should be implemented in cars. A vital component of those safety systems is the detection and prediction of motion of pedestrians. This is needed to prevent impending collisions. We utilize computer vision using high resolution video to predict where a pedestrian is likely to move in near future and whether a collision is imminent.

 

Pedestrian Detection and Feature Extraction
Pedestrians are detected in high resolution images. Based on a part wise detection, the gaze direction of a pedestrian is estimated using machine learning techniques. Also, other relevant features like the motion direction and speed are determined using tracking. 

 

Pedestrian Prediction
The extracted features are used as an input to a prediction stage. In this step, the history of the tracked features as well as information on the surrounding are used to determine possible future trajectories.