Real-time pedestrian recognition at night based on far infrared image sensor
Eunjin Choi, Wanjae Lee, Kanghoon Lee, Jaekwang Kim, Jinhak Kim · 2016
The damage of the accident between a pedestrian and a vehicle is most serious in the kind of traffic accidents. According to the statistics, 38% of road fatalities occur in an accident between a pedestrian and a vehicle, and the night accident is accounted for 64% in that number. This paper proposes pedestrian recognition algorithm with the far-infrared image sensor mounted vehicle at night time. We propose recognition algorithm with noble features which are Local Binary Pattern Haar-like (LBP-Haar_like), Advanced Histogram Oriented Gradient-Local Binary Pattern_histogram (adv_HOG- LBP _histogram) features. The features are extracted from big database (DB) using Adaptive Boosting (ada-boost) classification. The experimental results show that the proposed algorithm can detect and track pedestrian with 97% accuracy at average 20 frames per second.