Monocular vision-based object recognition for autonomous vehicle driving in a real driving environment

Nowadays, many attentions have been devoted to autonomous vehicles because the automation of driving technology has a large number of benefits, such as the minimization of risks, the improvement of mobility and ease of drivers. Among many technologies for autonomous driving, road environmental recognition is one of the key issues. In this paper, we present the test results of various object detection algorithms using single monocular camera for autonomous vehicle in real driving conditions. The vision recognition system tested in this paper has three main recognition parts: pedestrian detection, traffic sign and traffic light recognition. We use Histogram of Gradients (HOG) features and detect the pedestrians by Support Vector Machine (SVM). Also features of traffic signs are extracted by Principal Components Analysis (PCA) and canny edge detection is used for traffic lights. These two signals are classified by Neural Network (NN). Algorithms that we tested are implemented in General-Purpose computing on Graphics Processing Units (GPGPU). We show the effectiveness of these methods in real-time applications for autonomous driving.