Detection and Tracking Method of Multi-Pedestrian using Faster R-CNN and Kernelized Correlation Filter
The multi-pedestrian tracking technology used for autonomous driving and city congestion measurement should
be able to work in real time. The pedestrian tracking technique detects human from the image and tracks by sensing motion.
Typical human detection is done using the HOG characteristics which requires a large amount of computation because the
entire image is scanned from the input image to the windows of various sizes in order to detect human of various sizes.
However, the pedestrian detection based on the Faster R-CNN learns the object and background areas of the image through
the Region Proposal Network (RPN) to detect pedestrian, its processing time is much quicker. In this paper, we propose a
multi-pedestrian detection and tracking method that combines GPU-based Faster R-CNN with Kernelized Correlation Filter.
As a result, it was confirmed that pedestrian detection and tracking can be performed about 20% faster than the conventional
method, and the accuracy is improved by about 12%.
Keywords - Pedestrian Tracking, Faster R-CNN, Human Detection, Kernelized Correlation Filter