A SysML-based Design and Development of Stereo Vision System with Pose and Velocity Estimation for Cooperative Automated Vehicles

Cooperative automated vehicles must perceive the environment accurately and have precise information about the leading vehicle's pose and velocity. This paper presents a SysML-based approach to design and develop a stereo vision-based perception system in an urban platooning scenario. It detects objects, lane markers, and free space in front of the follower vehicle using deep neural networks and computes the lead vehicle's relative pose and velocity. The relative pose is estimated using a geometric model-based pose estimation algorithm. The relative velocity is estimated from the change in pose within a known time. The lead vehicle's relative pose and velocity control the follower vehicle to follow the lead vehicle autonomously. Also, it displays the lead vehicle's pose and velocity information in a meaningful way on the in-vehicle display of the follower vehicle. The proposed system uses a custom-built automotive-grade stereo camera as input and runs on an automotive-grade embedded platform. It is tested on a simulation and prototype cooperative automated vehicle research platform. The evaluation results demonstrate that the proposed system operates in real-time and is suitable for cooperative automated vehicles.