Ground estimation and point cloud segmentation using SpatioTemporal Conditional Random Field

Abstract

Whether it be to feed data for an object detectionand-tracking system or to generate proper occupancy grids, 3D point cloud extraction of the ground and data classification are critical processing tasks, on their efficiency can drastically depend the whole perception chain. Flat-ground assumption or form recognition in point clouds can either lead to systematic error, or massive calculations. This paper describes an adaptive method for ground labeling in 3D Point clouds, based on a local ground elevation estimation. The system proposes to model the ground as a Spatio-Temporal Conditional Random Field (STCRF). Spatial and temporal dependencies within the segmentation process are unified by a dynamic probabilistic framework based on the conditional random field (CRF). Ground elevation parameters are estimated in parallel in each node, using an interconnected Expectation Maximization (EM) algorithm variant. The approach, designed to target high-speed vehicle constraints and performs efficiently with highly-dense (Velodyne-64) and sparser (Ibeo-Lux) 3D point clouds, has been implemented and deployed on experimental vehicle and platforms, and are currently tested on embedded systems (Nvidia Jetson TX1, TK1). The experiments on real road data, in various situations (city, countryside, mountain roads,...), show promising results.

Ground Segmentation Algorithm for Sloped Terrain and Sparse LiDAR Point Cloud

“…Future research activities will be conducted on two different fronts: (i) real-time implementation, which is already in development with promising results using parallel programming with GPUs; and (ii) the exploration of temporal dependencies, which may lead to a more comprehensive description of the environment and thus a better estimation. Works such as [29] will be used as baseline.…”

“…However, its outcome is dependant of the single height condition which, as stated previously with elevation maps, struggles with overhanging structures and lack of points. Other representative techniques with higher order inference algorithms can be found, for example, in [27] which models the point cloud as a MRF, in [28], which uses Gaussian Process regression (GP), or in [29], which uses Conditional Random Fields (CRF) and extends the spatial dependencies to spatio-temporal dependencies.…”

“…Despite the different field of application, some authors have used airborne LiDAR approaches in autonomous vehicles applications. For example [29], mentioned above, was inspired by [32], which proposed in turn a new hybrid CRF model to extract Digital Terrain Model (DTM) and [33] applied the Cloth Simulation Filter (CSF) [34] to detect rocks on the road.…”

GndNet: Fast Ground Plane Estimation and Point Cloud Segmentation for Autonomous Vehicles

“…Occupancy grids are 2D spatial maps of the environment around the vehicle which can be constructed by processing the LiDAR point cloud data. Typical steps that precede the occupancy grid generation are the estimation of the ground plane and the segmentation of the ground points [3].…”

“…Currently, no public dataset is available with an annotated ground elevation map. To overcome this difficulty, we propose two approaches: morphological operations and a CRF-based [3] approach to obtain the ground-truth elevation map from the SemanticKITTI dataset [7], [8].…”