Pedestrian detection based on incremental learning

Abstract

Pedestrian detection is a hot topic in computer vision and pattern recognition. Existing pedestrian detection methods face new challenges in the background of big data, e.g., heavy burdens on computing and memory. To solve these problems, in this paper, we propose a pedestrian detection framework based on incremental learning. Compared with existing pedestrian detection frameworks, it costs much less time and memory. In addition, the performance of our framework is very close to the one which uses all training samples at once. Furthermore, with more new training samples, the performance can be enhanced continually with little time and memory, showing the potential in practical applications. © 2013 Springer-Verlag Berlin Heidelberg.