Vehicle Pose Estimation using Optical Flow for Inter-Vehicle Distance Compensation in Dynamic Scenes

Stable inter-vehicle distance estimations are required in dynamic scenes as they are essential for advanced driver-assistance systems (ADAS) and high-level autonomous driving. However, instantaneous changes in the pose of a vehicle in the dynamic scene may cause errors in the estimation of the inter-vehicle distance, which can lead to traffic accidents and also affect obstacle avoidance performance. In this study, the optical flow motion vector is histogrammed to extract a representative value, and the correlation between the representative value and inertial measurement unit (IMU) data is confirmed to verify the accuracy in vehicle pose estimation based on the optical flow motion vector. Then, an algorithm is proposed for improving the vehicle distance estimation accuracy in real-time ground compensation. Experiments have been conducted using an autonomous vehicle on the road with speed bumps, and the data from the IMU sensor installed in the vehicle and LiDAR detection results are used as the ground-truth data. The pose estimation with the proposed method shows an average error of 0.0029 rad, and the distance correction error can be reduced from 2.73 m (14.8%) to 0.86 m (4.6%), demonstrating the possibility of accurate real-time inter-vehicle distance correction. © 2022, Institute of Control, Robotics and Systems. All rights reserved.