Detecting pedestrians in still images using learned shape features

The problem of detecting pedestrians in images has received much attention from the computer vision community because of its variety of applications. This problem can be considered as a two-class classification problem by labeling windows cropped from the images as pedestrians or non-pedestrians. We present two novel methods for detecting pedestrians in still images. The first method uses coarse shape cues, and is based on a likelihood ratio test. Likelihoods for shape descriptors on pedestrian and non-pedestrian images are obtained using kernel density estimation. In the second approach, we introduce a new method for learning local discriminative features from training examples, and use them for object classification. This method uses two folds of the AdaBoost classifier, first for feature creation and second to train the final classifier. The quantitative results show that the performance of this method is better than the state of the art pedestrian detector.