Reconstruction of rigid body models from motion distorted laser range data using optical flow

Abstract

Abstract-The setup of tilting a 2D laser range finder up and down is a widespread strategy to acquire 3D point clouds. This setup requires that the scene is static while the robot takes a 3D scan. If an object moves through the scene during the measurement process and one does not take into account these movements, the resulting model will get distorted. This paper presents an approach to reconstruct the 3D model of a moving rigid object from the inconsistent set of 2D measurements by the help of a camera. Our approach utilizes optical flow in the camera images to estimate the motion in the image plane and point-line constraints to compensate the missing information about the motion in depth. We combine multiple sweeps and/or views into to a single consistent model using a point-to-plane ICP approach and optimize single sweeps by smoothing the resulting trajectory. Experiments obtained in real outdoor scenarios with moving cars demonstrate that our approach yields accurate models.

Velocity and shape from tightly-coupled LiDAR and camera

“…ICP and Kernel Correlation (KC). To compare with multimodal trackers, we use the C++ implementation 1 of Any-time Tracker (Held et al, [10]), and implement the method of Ilg et al, [29]. In order to tune hyperparameters in the model we use 5% of sequences in each dataset for training.…”

“…Non-parametric object tracking: Several methods align tracked object point clouds with new segmented scans, then append the segment to and enrich the 3D model [28], [29], [10]. Wyffels and Campbell [28] lay a probabilistic method to estimate motion and shape for a single extended object using simulated LiDAR data, but do not experiment with realworld datasets and do not add camera.…”

“…Scene flow using LiDAR and Camera: Ilg et al, [29] employ LiDAR and Camera, mounted on a stationary platform, to estimate pose and reconstruct the 3D model of a moving object. They project 3D laser points onto the image, and constrain the 3D motion based on precomputed optical flow at the projected pixel.…”

“…We employ velocity prior term as in (24) to enforce a smooth motion. We benchmark our results with ICP coupled by Kalman filtering, Any-time Tracker of Held et al, [10] and the method proposed by Ilg et al, [29]. For this experiment, we use KITTI Raw [36] dataset which consists of color images and high-resolution Velodyne HDL-64 point clouds synchronized with a 10Hz sampling rate.…”

On the optical flow model selection through metaheuristics

“…The literature is wide, being some very recent works related to optical flow estimation using Laplacian mesh structures [1], total generalized variation [2], probabilistic motion detection [3], and as an optimization problem in a high-dimensional motion field [4], just to name a few. The importance of optical flow estimation can be evidenced in image segmentation [5], rigid object reconstruction [6], cell tracking [7], video stabilization [8], among others. Some parallel-based implementations can be found in [9][10][11] as well.…”