3D Vehicle Pose Estimation from an Image Using Geometry

Abstract

3D pose estimation is a quite fascinating problem in the computer vision field, which aims to get the 3D orientation of the object based on the 2D image and can be challenging to find the solution. These solutions are mostly used in the autonomous industry to properly detect cars’ orientation on street scenes. Most solutions and publicly available datasets refer to only one axis orientation (axis parallel to the road), while two others are set to zero. In this paper, we propose a new way of representing a 3D object orientation that is not limited only to the autonomous industry. Our proposal uses Euler angle representation and two coordinate systems (one from the camera and one from the detected object) to define angle orientation for each axis in 3D space. Because this approach is supposed to be used in creating datasets and solutions in the deep learning field, there are some restrictions to be considered to properly train your solution. To better represent the proposal of 3D orientation and its purpose, it is necessary to visualize the bounding box around the detected object, which will follow its orientation. The visualization part is being done using computer-generated images (CGI), while the algorithmic part uses quaternions instead of Euler angles for the rotation of the 3D bounding box.