Pedestrian, bike, motorcycle, and vehicle classification via deep learning: Deep belief network and small training set
Yen-Yi Wu, Chun-Ming Tsai · 2016
In traffic monitoring environments where light changes a lot, classifying pedestrians, bikes, motorcycles, and other vehicles quickly is indeed a big challenge. For instance, pedestrians are of variable sizes, bikes of different styles, motorcycles of different shapes, and vehicles of different types. Because of these variations and can influence the classification results for these four categories. Recently, Deep Learning has often been used in object classification with reasonably good results, so interests in researching it for new applications have been aroused. However, Deep Learning is seldom used in researches of classifying pedestrians, bikes, motorcycles, and other vehicles. In this paper, Deep Belief Networks (DBN) of Deep Learning is applied to distinguish the above-mentioned four categories. The proposed DBN methods only used 1,000 image training set and could achieve a higher accuracy of classification rate.