Abstract: In this paper, aiming at the defects of existing detection method, being unable to effectively distinguish two types of object: pedestrians and cyclists, a concurrent pedestrian and cyclist detection method is proposed based on deep neural network (DNN), while in view of the problems of frequent undetection and false detection in codetection of pedestrian and cyclist in road environment, the poor detection results for smalldimension targets and the complex and changeable background environment, several DNN modification schemes like difficult example extraction, multilayer feature fusion and multitarget candidate region input are devised to realize concurrent pedestrian and cyclist detection. The results of tests on the public database of pedestrian and cyclist show that the method proposed achieves high identification rate of pedestrian and cyclist and can distinguish each other, with its effectiveness validated.

Key words: pedestrian and cyclist detection, deep neural network, feature fusion, candidate region proposal