Vehicle recognition system based on customised HOG for automotive driver assistance system

Abstract

In the last decade, advanced driver assistance systems (ADAS) made enormous progress. However, obstacle recognition tasks remain a challenge. In this paper, an optimisation vehicle detection system based on a customised histogram of oriented gradients (HOG) was presented and investigated to achieve an accurate vehicle recognition system. Our contribution in this work can be summarised in two fundamental points. First, a re-optimisation of the standard HOG parameters was made to get the best results for the car detection. Secondly, an amplification factor was distributed for each bin weight according to its contribution in the extracted car-features. Our studies using a linear support vector machine (SVM) classifier in MATLAB and heterogeneous databases of vehicle and non-vehicle images were made to achieve an excellent recognition rate that outperforms other similar approaches.