A Two-Stage Optimal Detection Algorithm Research for Pedestrians in Front of the Vehicles

In this paper, a two-stage optimal detection algorithm is presented for pedestrians in front of the vehicles. It uses the idea of combing the coarse-grain and fine-grain to effectively classify and filter. First, it uses the combination of Color Self-Similarity features based on rectangular block summing and AdaBoost classifier based on greedy strategy to coarse-grained screen the pedestrian detection window, then it uses the combination of HOG feature and libsvm classifier to fine-grained confirm the previous screened pedestrian detection window, Finally, the target windows is integrated by the greedy strategy. The AdaBoost classifier's training time is theoretically shorten to the 1/T time of original algorithm. With the training process, The Color Self-Similarity features shorten to 250 dimensions by the feature selection. Then, the method makes full use of the image information and ensures the detection accuracy.