Abstract:

In driver assistance systems,real-time and accurate pedestrian detection is required. In this paper,by taking advantage of the low complexity of a knowledge-based detection method,a nighttime pedestrian detection method for monocular far-infrared video data is proposed based on the probabilistic template matching. In this method,according to the distribution of intensity in pedestrian samples,a local dual threshold segmentation algorithm is adopted to extract the candidate regions that may contain pedestrians. Then,multi-scale probabilistic templates are established based on the moving directions of pedestrian samples and are employed to recognize the potential pedestrians from the candidate regions. The establishment mode of probabilistic templates alleviates the large withinclass variability of pedestrian samples,thus improving the induction abilities of the probabilistic templates for the appearance of pedestrians. In order to further improve the detection accuracy,the probabilistic template matching is integrated with the object-tracking algorithm,which results in more robust final decision through the multi-frame validation. Experimental results show that the proposed method can realize a real-time pedestrian detection with a low computation cost; and that it achieves a detection rate of more than 90% at the false alarm rate of less than 10% on suburban scenes while a detection rate of about 75% at the false alarm rate of about 22% on urban scenes.

Key words: driver assistance systems, pedestrian detection, probabilistic template, object tracking, far-infrared video