Frame Level Difference (FLD) Features to Detect Partially Occluded Pedestrian for ADAS

Abstract

Computer vision-based technologies take a significant position in the enhancement strategies of automation industries by identifying and tracking of persons on the road mainly for Advanced Driver Assistant System (ADAS). Features are an important proposition in decision of accuracy during the identification process of pedestrians. The selected features are very low in quality because of the poor lighting conditions, sensors used, occlusion and amount of distortions present in the motion video. A unique Frame level Difference (FLD) features is proposed and will extract the features by finding the difference between the adjacent frame and retaining the noticeable differences. The proposed one supports recognizing the pedestrians in the presence of occlusion. Experiments are carried out with the standard Caltech pedestrian dataset and the results demonstrated by using combination of proposed features with other existing one to improve the detection accuracy. Also avoids the greater number of false positives and a larger proportion of miss rates.