ADAPTIVE BAYESIAN COMBINATION OF FEATURES FROM LASER SCANNER AND CAMERA FOR PEDESTRIAN DETECTION

This paper presents a probabilistic approach to combine the information extracted from a laser scanner and video camera, in order to detect the presence of pedestrians along the vehicle path. Our method consists of extracting the objects from the laser scanner data, classifying them in the both sensors and combining the classifiers decisions using the Bayes formula. In the laser scanner space, the classification is based on heuristic rules about the pedestrians outlines and their kinematic constraints. The image region candidates detected are confirmed for the presence of pedestrians by a statistical learning algorithm, Adaboost. Our Bayesian classifier combines the results of each classification techniques by computing the probability of being a pedestrian for each detected object. The paper focus on the past knowledge integration which reduces the spurious detection effects; and an original way to include scale variation in vision-based classifier. The latter allows improving vision based classification of far or small pedestrians. Experimental results carried out illustrate the robustness of our approach even in cluttered road scene.