Enhancing Pedestrian Detection Using Context Information

Abstract

Detecting pedestrians among other objects in a digital image is a relevant task in the field of computer vision. This paper presents a method to improve the performance of a pedestrian detection algorithm using context information. A neural network is used to classify the region below pedestrian candidates as being floor or non-floor. We assume that a pedestrian must be standing on a floor area. This scene context information is used to eliminate some of the false-positive pedestrian candidates, therefore improving detector precision. The neural network uses 10 feature channels extracted from the original image to perform the region classification. This method may be used along with a large family of pedestrian-detecting algorithms. We used the ACF-LDCF algorithm to perform the tests in this research. The result shows that this method is very effective. We achieve a gain of 7% in ACF-LDCF algorithm performance on the Caltech pedestrian benchmark.