A cost-effective computer vision-based vehicle detection system

Vehicle detection plays an important role in the development of an autonomous driving system. Fast processing and accurate detection are two major aspects of generating the autonomous vehicle detection system. This paper proposes a novel computer vision-based cost-effective vehicle detection system. Here, a Gentle Adaptive Boosting algorithm is trained with Haar-like features to generate the hypothesis of vehicles. Haar-like feature generates hypotheses very fast but may detect false vehicle candidates. The support vector machine algorithm is trained with the histogram of oriented gradient features to filter out the generated false hypothesis. The histogram of oriented gradients descriptor utilizes the shape and outlines of the vehicles, hence detects vehicles more accurately. Haar-Likes features and histogram of oriented gradients features are organized to accomplish the aspects of autonomous driving. The performance of the proposed vehicle detector is evaluated for day time and night time captured images and compared with three different existing vehicle detectors. The average precision of the proposed system for day time captured image is 0.97 and for night time captured image is 0.94. The proposed system requires 15 times less training time as compared to the existing technique for the same number of image data and on the same CPU.