Development of Onboard Online Ensemble learning Function for Evolutionary Risk Recognition

摘要

Conventional recognition techniques require the consideration of various environmental factors such as time, place and weather in its development. To address this problem, an onboard online-learning system was developed to improve vehicle safety which automatically constructs a pedestrian detection algorithm. An onboard camera system captures images which are used as target information. Based on the images, the learning system produces several recognizers, calculates the optimal combination of the recognizers and integrates them. Our experiment results show that the detection ability of pedestrians has been improved with the new system.