One of the most frequent infractions on the road is the act of a vehicle crossing to the wrong side of the road to pass another vehicle traveling in the same direction. Automatic detection of this violation can be a challenging issue. Thus, we aim to develop a computer vision system to robustly detect forbidden overtaking observed from a fixed camera. Our approach is based on two main phases: Line Detection and Vehicle Detection. In this work, we focus on the vehicle detection stage. Here, features are extracted from the image and then classified using machine learning algorithms. In a first experiment, we constructed different models using features such as HOG (Histogram of Oriented Gradient), SURF (Speeded up Robust Features), Gabor filter and LBP (Local Binary Patterns), and machine learning classifiers as SVM (Support Vector Machines), kNN (k-Nearest Neighbor) and Decision Tree. Then, we merged the best descriptors to combine the advantages of their different robustness in order to build a strong vehicle detection model. The performances of all constructed models are evaluated on the GTI database. In a second experiment, illumination normalization techniques were applied to image database for the effective models. Thus, the first experimental results reveal that the combination HOG+ LBP+ Gabor Phase performs well with SVM. The second experimental results show that the Variational Retinex algorithm-based illumination correction provides a significant improvement in the detection rate.