Image Segmentation Technique for on Road Pedestrian Detection During Night Time

On-road pedestrians are becoming vulnerable due to the increasing road traffic. They are more vulnerable during night time due to low visibility and increasing risk due to driver's fatigue while driving. This work is aimed at developing a segmentation algorithm based on the dual-level adaptive thresholding approach, referenced from the available literatures. However the normal approaches show certain shortfalls especially for a near distance pedestrian. The work presented here explains the modifications to the dual thresholding algorithm to enhance the segmentation. Instead of fully dependent on the intensity of object as suggested in the referenced literature, the color properties of the object and saturation intensity is also used during preprocessing and segmentation. Further, rather than using 1-D odd window, 1-D even window is used to calculate the different thresholding levels. Enhancement in segmentation algorithm is extensively tested for number of scenarios with the end-to-end pedestrian detection system developed indigenously. This paper is concluded highlighting the results of improved segmentation output with modified algorithm. The paper is the authoritative source for the abstract.