A Study of On-board Fusion and Extraction Algorithm of Pedestrian under the Environment of CVIS

Due to its low accuracy of detection distance and occlusion,current on-board pedestrian recognition sys-tems are not suitable for implementation at curves and intersections.In order to enhance the warning effects of vehicle on-board system,this paper developed a fusion & extraction algorithm for pedestrian identification under the environment of cooperative vehicle infrastructure system (CVIS).Pedestrian is detected by roadside and on-board cameras,and pretrea-ted by a Kalman filter.The optimal position estimation of pedestrian is estimated through spatial and temporal alignment, fuzzy association and Kalman fusion.An actual vehicle experiment platform is set up to verify the effectiveness of the pro-posed fusion & extraction algorithm.Compared to errors from the pedestrian trajectory estimated before applying the fu-sion and extraction algorithm ,the maximum absolute error reduces by 5 0 .0 0 % and absolute average error reduces by 55.56% at the X direction .As to theY direction,the maximum absolute error reduces by 40.00%,and the absolute av-erage error reduces by 62.07%.Experimental results show that the proposed algorithm improves the detection accuracy of pedestrian trajectory and enhances the warning accuracy of the system significantly.