Real-time multiple pedestrian tracking framework using circulant structure of kernels

Pedestrian detection (PD) and tracking is one of the most prominent functionalities in ADAS (Advanced Driver Assistance Systems). These systems use a forward looking camera (FLC) to detect pedestrians and warn the driver in the host vehicle for the possibility of collision. In the past decade, multiple pedestrian tracking has been a significant and crucial area of research. This paper describes the implementation of multiple pedestrian tracking in a pedestrian detection framework. The focus of this tracking approach is to improve the accuracy of detection whilst achieving a low processing time. The proposed tracking framework uses the concept of tracking-by-detection approach with CSK (circulant structure of kernels). CSK combined with Fourier transform using Gaussian Kernels helps in achieving very fast learning and detection of the target. The proposed tracking framework achieves 4.5% increase in accuracy of detection.