Pedestrian detection: A benchmark

Abstract

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 perwindow 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-theart 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 www.vision.caltech.edu/Image Datasets/CaltechPedestrians/ 1 304 978-1-4244-3991-1/09/$25.00 ©2009 IEEE

Efficient Labelling of Pedestrian Supervisions

“…For evaluation and performance comparison, Recall-FPPI (False Positives Per Image) curves are used since they are commonly used in state-of-the-art object detection research [42,14]. Especially for pedestrian detection, Recall-FPPI curves provides information, at a glance, about the recall at various false positives per image.…”

Vision-Based System for Human Detection and Tracking in Indoor Environment

“…[7]) or with the connected component detected with the background subtraction. According to the results of human detection in a sliding detection framework, presented for example in [9], performances of these detectors do not seem sufficient to initialize the tracking for robust applications. It thus seems more judicious to use connected components obtained with background subtraction.…”

Enhancing Pedestrian Detection Using Context Information

“…Some of these strides are due to the challenges posed by image databases used for training and testing (Dollár et al, 2009). The challenges imposed by the databases are catalysts for progress in some fields of computational vision.…”

“…Extensive tests were applied along with the Caltech pedestrian database (Dollár et al, 2009). The Caltech-USA database is known as a benchmark in pedestrian detection algorithms and is widely used for training and testing.…”

“…Results show that the use of floor presence information can improve the accuracy of a pedestrian detection algorithm by up to 7%. The classifier was trained with a large number of example patches selected from images from the Caltech-USA database (Dollár et al, 2009). Positive examples were cropped from regions where pedestrians can walk: asphalt, sidewalk and grass.…”