An Efficient Technique for Night Time Vehicle Detection with Fusion Based Image Enhancement

Vehicle detection at night time is a challenging problem which deals with inadequately illuminated images of low contrast and reduced visibility. In this paper, an efficient region of interest (ROI) extraction approach which combines vehicle light detection and object proposals together with fusion-based score-level feature technique is proposed. The scorelevel multi-feature fusion method involves seven complementary features such as difference of Gaussian (DOG) , Scale Invariant Feature Transform (SIFT) ,local binary pattern (LBP), histogram of oriented gradients (HOG), fourdirection features (FDF), HSV color histogram and edge HOG which represent different attributes of the vehicle and demonstrate different accuracy for vehicle recognition. In addition Elastic Edge Boxes is utilized on enhanced images to generate original object proposals. Eventually, the versatile AdaBoost classifiers are used to classify the corresponding features extracted from the ROIs for improving the accuracy of nighttime vehicle detection. Keyword: Region of interest, Elastic edge boxes, Score Vector Machine, AdaBoost.