Evaluation of Keypoint Descriptors Applied in the Pedestrian Detection in Low Quality Images

Pedestrian detection is a basic task in video surveillance for systems as of driver assistance systems, tracking pedestrian, detection of anomalous behavior, among others. Local features detectors and descriptors are widely used in many computer vision applications and several methods have been proposed in recent years. Performance evaluation of them is a tradition in computer vision; however, there is a gap comparative of traditional keypoint descriptors like SIFT, SURF and FAST against recent and novel local feature extractors such as ORB, BRISK and FREAK in low quality images, because when the number of pixels representing an object is low, the ability to recognize the object is reduced. This article aims to present a systematic and comparative study of the performance these local features detectors and descriptors in pedestrian detection in four real databases, all in an urban environment.