Bicyclist recognition and orientation estimation from on-board vision system

Abstract

Bicyclists move at speed equivalent to a slowly moving vehicle, and sometimes share the road with vehicles in urban environments. Thus, bicyclists take more challenge for safe-driving compared to pedestrians. Therefore, accident avoidance system is expected to recognize the type of road user, and then can perform further behavior analysis for risk assessment based on the type of road user. Our contribution to this tendency consists of a method to distinguish bicyclists from pedestrians, and reliably estimate the bicyclists' head orientation and body orientation from image sequences taken by an on-board camera. The output of the proposed method can be used for risk assessment.