A Trainable System for Object Detection
International Journal of Computer Vision - special issue on learning and vision at the center for biological and computational learning, Massachusetts Institute of Technology
Distinctive Image Features from Scale-Invariant Keypoints
International Journal of Computer Vision
Histograms of Oriented Gradients for Human Detection
CVPR '05 Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) - Volume 1 - Volume 01
Fast Human Detection Using a Cascade of Histograms of Oriented Gradients
CVPR '06 Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Volume 2
Monocular Pedestrian Detection: Survey and Experiments
IEEE Transactions on Pattern Analysis and Machine Intelligence
Object Detection with Discriminatively Trained Part-Based Models
IEEE Transactions on Pattern Analysis and Machine Intelligence
ViBe: A Universal Background Subtraction Algorithm for Video Sequences
IEEE Transactions on Image Processing
| Hi-index | 0.00 |
Pedestrian detection is a fundamental problem in video surveillance. An overwhelming majority of existing detection methods are based on sliding windows with exhaustive multi-scale scanning over the whole frame images which can achieve good accuracy but suffer from expensive computational cost. To reduce the complexity significantly while keeping high accuracy, in this paper, we propose an effective and efficient pedestrian detection method based on sliding windows with well-designed multi-scale scanning over candidate regions instead of whole frames. The candidate regions can be obtained through three main steps: (1) foreground extraction by using a fast background subtraction model to remove large number of static regions since pedestrians are usually keeping moving; (2) region merging and filtering through clustering foreground pixels to avoid over-partitioned or too large regions of non-pedestrian; (3) well-designed multi-scale scanning by exploiting the size information of current region to avoid useless scales. Therefore, through utilization of motion and size information, we can not only speed up the detection through reducing large number of windows, but also improve the accuracy of detection through eliminating many false positive regions. Our experiments on two public datasets have verified that our method outperforms the state-of-the-art methods in both speed and accuracy of detection.