Modeling Traffic Scenes for Intelligent Vehicles Using CNN-Based Detection and Orientation Estimation

Object identification in images taken from moving vehicles is still a complex task within the computer vision field due to the dynamism of the scenes and the poorly defined structures of the environment. This research proposes an efficient approach to perform recognition on images from a stereo camera, with the goal of gaining insight of traffic scenes in urban and road environments. We rely on a deep learning framework able to simultaneously identify a broad range of entities, such as vehicles, pedestrians or cyclists, with a frame rate compatible with the strong requirements of onboard automotive applications. The results demonstrate the capabilities of the perception system for a wide variety of situations, thus providing valuable information to understand the traffic scenario.