LIMO: Lidar-Monocular Visual Odometry

Abstract

Higher level functionality in autonomous driving depends strongly on a precise motion estimate of the vehicle. Powerful algorithms have been developed. However, their great majority focuses on either binocular imagery or pure LIDAR measurements. The promising combination of camera and LI-DAR for visual localization has mostly been unattended. In this work we fill this gap, by proposing a depth extraction algorithm from LIDAR measurements for camera feature tracks and estimating motion by robustified keyframe based Bundle Adjustment. Semantic labeling is used for outlier rejection and weighting of vegetation landmarks. The capability of this sensor combination is demonstrated on the competitive KITTI dataset, achieving a placement among the top 15. The code is released to the community.

Lidar-Monocular Visual Odometry with Genetic Algorithm for Parameter Optimization

“…The whole system takes significant time since we are running 50 * 50 = 2500 LIMO evaluations to determine the best parameters. The next section shows our experiments with individual and combined sequences, with and without the GA. Our results show that the GA-LIMO performs better than the results of LIMO [7].…”

“…Closer to our research, GAs have been applied to early SLAM optimization problems [32], mobile localization using ultrasonic sensors [33] [34], and in deep reinforcement learning [35]. This provides good evidence for GA efficacy on localization problems, and our main contribution in this paper is a demonstration of smaller translation error when using a GA to tune LIMO parameter values compared to the stock LIMO algorithm [7]. Our experiments show that translation error is non-linearly related to LIMO parameters, that is, translation error can vary nonlinearly based on the values of the LIMO's parameters.…”

“…This dataset has rural, urban scenes along with highway sequences and provides gray scale images, color images, LIDAR point clouds and their calibration. Most LIMO configurations are as in [7]. In our work, we focus on two sequences in particular: sequence 01 and 04.…”

“…Recent developments make use of offline VSLAM for mapping and localization [4]- [6]. Figure 1 illustrates the structure of the VSLAM pipeline [7]. Algorithms such as ROCC [8] and SOFT [9] rely on preprocessing and feature extraction which is in very contrast to most of the methods that obtain scale information from a camera placed at a different viewpoint [6], [10]- [12].…”

“…The input is temporal sequence of images, and the system outputs a sparse reconstruction of the observed environment and the camera poses. [10] [21] [22] [7]. In this work, LIMO does not perform loop closure.…”