Loose coupling visual-lidar odometry by combining VISO2 and LOAM

Odometry is a very popular research direction in recent decades and plays a very critical role in the autonomous navigation system. VISO2, as a relatively robust and mature visual odometry algorithm, becomes unreliable when distinct visual features are insufficiently or when the texture of the environment is repetitive. Even if not in such a challenging environment, the drift is still a big problem for the practical application. LOAM, a lidar odometry algorithm, has been leading the way for the performance on KITTI data set. In fact, it also has limitations in shape features not prominent environment, like a highway with few buildings around. Considering that the current mobile unmanned platforms (e.g., unmanned vehicles) are almost equipped with both lidar and cameras, this paper proposes a loose coupling visual-lidar odometry by combining VISO2 and LOAM to achieve the advantages of both camera and lidar and reduce the number of restricted scenarios. The algorithm proposed in this paper is tested on the raw data set of KITTI. The average translation error is about 1.37% on 10 sequences, and the accuracy is improved with respect to LOAM with the same parameters. With the help of VISO2, the algorithm also performs well in shape features not prominent environment (e.g., highway), and from the experiment results, we confirmed the efficient strategy to combine VISO2 and LOAM to adapt to challenging environment and to keep the accuracy of the algorithm.