2D laser based road obstacle classification for road safety improvement

Abstract

Vehicle and pedestrian collisions often result in fatality to the vulnerable road users (VRU), indicating a strong need of technologies to protect such persons. Laser sensors have been extensively used for moving obstacles detection and tracking. Laser impacts are produced by reflection on these obstacles which suggests that more information is available for their classification. This paper proposes a new system to address this issue. We introduce the design of our system that is divided in three parts : definition of geometric features describing road obstacles, multiclass object classification from an adaboost trained classifier and track class assignment by integrating consecutive classification decision values. During this study, we show how specific features adapted to urban obstacles enhance the state of the art method for person detection in 2D laser data. Hence, in this paper, we evaluate usefulness of each feature and list the best ones. Moreover, we investigate the influence of laser height for each class showing that classification performance depends on the sensor position. Finally, we tested our system on some laser sequences and showed that it can estimate the class of some road obstacles around the vehicle with an accuracy of 87.4%.

An Efficient L-Shape Fitting Method for Vehicle Pose Detection with 2D LiDAR

“…Another solution is similar to the approach widely used in computer vision by extracting hand-crafted features and training classifiers. Image-based object detection is very popular in current autonomous driving research, such as road obstacles detection [6], mobility aids [7] and vehicle detection [8]. However, the sparse point data from a 2D Li-DAR are usually insufficient for reliable object identification using this kind of method within a single scan.…”

A Cooperative Car-Following/Emergency Braking System With Prediction-Based Pedestrian Avoidance Capabilities

“…As highlighted in [11] the adjacent property can be found false in the case of partial occlusion or objects with holes but for the current platoon application this limitation was found acceptable. The classification used is very simple but efficient and stable enough when used in conjunction with the tracking strategy that temporally refines classification as in [23].…”

DROW: Real-Time Deep Learning-Based Wheelchair Detection in 2-D Range Data

“…Merdrignac et al [18] also detect cars, bicyclists and static road obstacles in 2D range data. They hypothesize that more 6 https://arxiv.org/abs/1603.02636v1 information can be extracted from range data than done before and aim to achieve this by designing a large set of hand-crafted features.…”

“…While early detection methods used simple heuristics such as fitting lines and circles [35], in the past few years hand crafted features, coupled with learned classifiers, have dominated laser based detection. Within this paradigm, a range of successful people [1], [30], [20], mobility aid [34] and road obstacle [18] detectors have been developed to support mobile robot navigation [8] and autonomous driving [28]. Even though the aforementioned models obtain respectable results, the general consensus seems to be that the information provided by 2D range data is not sufficient to reliably perform detection in a single scan, leading to approaches relying on sensor fusion [30], multilayered sensor setups [20], [29], or temporal integration of information by tracking.…”