Pedestrian detection: A benchmark

Pedestrian detection is a key problem in computer vision, with several applications including robotics, surveillance and automotive safety. Much of the progress of the past few years has been driven by the availability of challenging public datasets. To continue the rapid rate of innovation, we introduce the Caltech Pedestrian Dataset, which is two orders of magnitude larger than existing datasets. The dataset contains richly annotated video, recorded from a moving vehicle, with challenging images of low resolution and frequently occluded people. We propose improved evaluation metrics, demonstrating that commonly used per-window measures are flawed and can fail to predict performance on full images. We also benchmark several promising detection systems, providing an overview of state-of-the-art performance and a direct, unbiased comparison of existing methods. Finally, by analyzing common failure cases, we help identify future research directions for the field.

1. Introduction

Detecting people in images is a problem with a long history [37], [13], [35], [27], [16], [41], [23], [5]; in the past two years there has been a surge of interest in pedestrian detection [6], [9], [11], [18], [20], [21], [25], [30], [32], [33], [36], [38], [42]. Accurate pedestrian detection would have immediate and far reaching impact to applications such as surveillance, robotics, assistive technology for the visually impaired, content based indexing (e.g. Flickr, Google, movies), advanced human machine interfaces and automotive safety, among others. Automotive applications [12], [14], [34] are particularly compelling as they have the potential to save numerous lives [39].