What could move? Finding cars, pedestrians and bicyclists in 3D laser data

By Dominic Zeng Wang, Ingmar Posner, Paul M. Newman. 2012

120 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Environmental Engineering, Ocean Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Environmental Engineering

In Ocean Engineering

Citations in the corpus

3D fully convolutional network for vehicle detection in point cloud (2016, background): “[16] also suggest it possible to predict 3D object localization by 2D CNN network on range scans.”
Vehicle Detection from 3D Lidar Using Fully Convolutional Network (2016, background): “More delicate segmentation can be obtained by forming graphs on the point cloud [32, 14, 21, 29, 30].”
Voting for Voting in Online Point Cloud Object Detection (2015, methodology): “[20] take a related approach, where a laser scan is first oversegmented into object parts, and SVM classifiers are trained to separate the segments into either foreground that belongs to objects of interest or the background.”
Detection and tracking of pedestrians and vehicles using roadside LiDAR sensors (2019, methodology): “The visual-based methods, such as the mixtures of Gaussians (MoG) method (Stauffer and Grimson, 2000), statistical background modeling (Wang et al, 2012) and convolutional neural network deep learning method (Sakkos et al., 2017; Babaee et al., 2018) cannot be used since the LiDAR data are point…”
A Survey on 3D Object Detection Methods for Autonomous Driving Applications (2019, unknown): “, graph-based segmentation [10] and voxel-based clus-”
Lidar for Autonomous Driving: The Principles, Challenges, and Trends for Automotive Lidar and Perception Systems (2020, methodology): “From the arsenal of ML, plenty of ML tools such as naive Bayes [25], support vector machines (SVMs) [21], [24], [28], k-nearest neighbor (NN) [26], random forest (RF), and gradient boosting trees [29] can be applied.”
Opening up Open World Tracking (2021, methodology): “Early methods in vision-based tracking [27,59,85] and robotic perception [54,76] utilized class-agnostic, bottom-up segmentation as a tracking cue, e.g ., based on LiDAR point cloud clustering [53, 79] or background modeling and foreground grouping [32,73,85].”
Distraction suppression for vision-based pose estimation at city scales (2013, background): “[25] to postprocess our prior maps and remove most of the dynamic objects in the scene, which include pedestrians, vehicles, and cyclists.”
Model-free detection and tracking of dynamic objects with 2D lidar (2015, background): “Euclidean Minimum Spanning Tree (EMST)-based clustering algorithms have a long history (Zahn, 1971; Asano et al., 1988; Wang et al., 2012), and is known for being capable of detecting clusters with irregular boundaries (Zahn, 1971).”
Performance of global descriptors for velodyne-based urban object recognition (2014, methodology): “Wang et al. [14] concatenated a SI computed at the centroid, shape distributions, shape factors, and Bounding Box as the global descriptors for the dynamic urban object recognition.”
Multi-scale object candidates for generic object tracking in street scenes (2016, methodology): “[16] use a minimal-spanning-tree clustering approach to extract 3D object proposals from LIDAR data and then classify them into background, bicyclist, car, or pedestrian.”
SqueezeSeg: Convolutional Neural Nets with Recurrent CRF for Real-Time Road-Object Segmentation from 3D LiDAR Point Cloud (2017, background): “This paradigm, despite its popularity [2], [3], [4], [5], has several disadvantages: a) Ground segmentation in the above pipeline usually relies on hand-crafted features or decision rules – some approaches rely on a scalar threshold [6] and others require more complicated features such as surface no”