Pedestrian detection via sensors even more reliable in the future?

(16-11-2021) Pedestrian detection is important in computing applications such as autonomous driving and video surveillance. In his PhD, Zhixin Guo investigates how pedestrian detection can be made even more reliable in the future.

In the past decade, the accuracy and reliability of modern pedestrian detectors have already improved significantly. Yet, more detailed studies are still needed to address a number of challenging situations.

The occlusion problem is an inevitable problem for pedestrian detection.

"In urban streets, pedestrians are not always easily visible if they are completely or partially hidden behind neighboring pedestrians or vehicles. In that case, the error rate of ordinary pedestrian detectors increase greatly," Zhixin Guo explains.

Another problem with pedestrian detection is the low contrast in some situations.

"On cloudy days or at night, the poor lighting condition will make it very difficult for the detector to recognize pedestrians," Zhixin continued.

The efficiency problem in data storage also appears to be a major concern for pedestrian detection. After all, modern detectors require high computational and storage costs.

"In my research, I sought solutions to the above problems so that pedestrian detection can become even more reliable and safe in the future," Zhixin concludes.

Read the entire PhD

-

PhD Title: Occlusion Handling and Multi-Modality Fusion for Efficient Pedestrian Detection

-

Contact: Zhixin Guo, Wenzhi Liao (promotor), Wilfried Philips (promotor), Peter Veelaert (promotor)

-

Editor: Jeroen Ongenae - Illustrator: Roger Van Hecke