Vision-based bicycle detection and tracking using a deformable part model and an EKF algorithm

Abstract

Bicycles that share the road with intelligent vehicles present particular challenges for automated perception systems. Bicycle detection is important because bicycles share the road with vehicles and can move at comparable speeds in urban environments. From a computer vision standpoint, bicycle detection is challenging as bicycle's appearance can change dramatically between viewpoints and a person riding on the bicycle is a non-rigid object. In this paper, we present a vision-based framework to detect and track bicycles that takes into account these issues. A mixture model of multiple viewpoints is defined and trained via a Support Vector Machine (SVM) to detect bicycles under a variety of circumstances. Each component of the model uses a part-based representation and known geometric context is used to improve overall detection efficiency. An extended Kalman filter (EKF) is used to estimate the position and velocity of the bicycle in vehicle coordinates. We demonstrate the effectiveness of this approach through a series of experiments run on video data of moving bicycles captured from a vehicle-mounted camera