A SVM-based classifier with shape and motion features for a pedestrian detection system

The most critical requirement of a pedestrian detection system is to quickly recognize pedestrians in an image. However, the huge number of candidate regions and the complexity of scenes usually make the recognition slow and unreliable. An efficient classifier is needed for a pedestrian detection system. In this paper, a decomposed SVM algorithm is used to train a classifier for pedestrian detection. The algorithm is stable and suitable for training a classifier with a large number of samples and the derived classifier is very efficient. Meanwhile, considering that our system is based on a single camera and the scenes are always complex, it is difficult to train a good classifier only with shape features. To solve these problems, we integrate shape information with motion information to compose a feature set and use it to train a classifier. Experiments show that our system based on this classifier works very well. Furthermore, we discuss the effect of applying motion features. With a proper percentage, motion features will be a good complement of the shape features in complex scenes. Comparison between application of shape features and application of both shape and motion features shows the advantage of our method.