Abstract:

It is rather difficult to detect pedestrians accurately in the traffic images stained by speckle noise and in- tensity distortions under complex illumination.In order to solve this problem and improve the accuracy and automa- tion level of information extraction from traffic images,a new pedestrian detection method,which is based on adap- tive pulse-coupled neural networks,is proposed.In the investigation,first,the ignition contribution values between the central nerve and its neighborhoods are determined according to the quasi-Euclidean distance between pixels.Then,a key control parameter named initial threshold is set by merging gray feature and neighborhood information.Finally,multi-strategy morphological modifications are performed on the initial detection results to obtain the final pedestrian information.Experimental results demonstrate that the proposed method greatly eliminates the impact of noise,well restrains the over-segmentation,and helps to obtain satisfactory detection results with good adaptability.

Key words: intelligent transportation, pedestrian detection, pulse-coupled neural networks, computer vision, adaptability