21 April 2016 Far-infrared pedestrian detection for advanced driver assistance systems using scene context
Author Affiliations +
Pedestrian detection is one of the most critical but challenging components in advanced driver assistance systems. Far-infrared (FIR) images are well-suited for pedestrian detection even in a dark environment. However, most current detection approaches just focus on pedestrian patterns themselves, where robust and real-time detection cannot be well achieved. We propose a fast FIR pedestrian detection approach, called MAP-HOGLBP-T, to explicitly exploit the scene context for the driver assistance system. In MAP-HOGLBP-T, three algorithms are developed to exploit the scene contextual information from roads, vehicles, and background objects of high homogeneity, and we employ the Bayesian approach to build a classifier learner which respects the scene contextual information. We also develop a multiframe approval scheme to enhance the detection performance based on spatiotemporal continuity of pedestrians. Our empirical study on real-world datasets has demonstrated the efficiency and effectiveness of the proposed method. The performance is shown to be better than that of state-of-the-art low-level feature-based approaches.
© 2016 Society of Photo-Optical Instrumentation Engineers (SPIE)
Guohua Wang, Guohua Wang, Qiong Liu, Qiong Liu, Qingyao Wu, Qingyao Wu, } "Far-infrared pedestrian detection for advanced driver assistance systems using scene context," Optical Engineering 55(4), 043105 (21 April 2016). https://doi.org/10.1117/1.OE.55.4.043105 . Submission:


Road surveillance using a team of small UAVs
Proceedings of SPIE (April 30 2009)
Invariant region descriptors for robust shot segmentation
Proceedings of SPIE (January 15 2006)
Automated counting of pedestrians
Proceedings of SPIE (September 15 1994)

Back to Top