Towards 3D object recognition via classification of arbitrary object tracks

By Alex Teichman, Jesse Levinson, Sebastian Thrun. 2011

198 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

How researchers see this paper

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

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

3D fully convolutional network for vehicle detection in point cloud (2016, methodology): “For the classification stage, research have been drawn to features including shape model [6], [13] and geometry statistic features [22], [27], [29], [31].”
Vehicle Detection from 3D Lidar Using Fully Convolutional Network (2016, background): “Teichman et al. [28] also encodes the object moving track information for classiﬁcation.”
Voting for Voting in Online Point Cloud Object Detection (2015, background): “[19, 18], for example, first segment input laser scans and then track the segments obtained in a sequence of scans.”
FusionNet: 3D Object Classification Using Multiple Data Representations (2016, neutral): “Classification of 3D objects has been proposed using hand-crafted features along with a machine learning classifier in [14], [25] and [1].”
Towards fully autonomous driving: Systems and algorithms (2011, background): “The following is a brief overview of the key points of this system; see [6] for details.”
Pedestrian detection combining RGB and dense LIDAR data (2014, background): “Except in [26] where the authors use HOG on 3D point cloud segments, to our best knowledge, there are no scientific work on pedestrian detection that use HOG directly on dense-depth maps.”
Laser-based segment classification using a mixture of bag-of-words (2013, background): “Especially self-driving cars need to distinguish reliably between static and dynamic objects, such as pedestrians, vehicles, and bicyclists, to ensure safe operation in crowded and even uncooperative inner-city traffic.”
VoxNet: A 3D Convolutional Neural Network for real-time object recognition (2015, methodology): “The input to our algorithm is a point cloud segment, which can originate from segmentation methods such as [12], [29], or a " sliding box " if performing detection.”
Research Advances and Challenges of Autonomous and Connected Ground Vehicles (2019, background): “In [35], [97], the shape of objects and the motion characteristics are utilized to identify pedestrians, bicyclists, and cars.”
SPNet: Deep 3D Object Classification and Retrieval using Stereographic Projection (2018, methodology): “Point cloud-based methods: While previous works often combine hand-crafted features or descriptors with a machine learning classiﬁer [11,32,4,9], the point cloud-based methods operate directly on point clouds in an end-to-end manner.”
Tracking-based semi-supervised learning (2011, methodology): “As in [21], we instead use the normalized discrete Bayes filter”
All Weather Perception: Joint Data Association, Tracking, and Classification for Autonomous Ground Vehicles (2016, background): “Examples include: improved tracking and classiﬁcation algorithms that ignored any segmentation or data association errors (Teichman and Thrun, 2011), performed evaluations on very limited types of scenarios such as tracking a large number of stationary cars or a small number of dynamic objects…”
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].”
What could move? Finding cars, pedestrians and bicyclists in 3D laser data (2012, methodology): “In particular, we make use of the Stanford Track Collection (STC) dataset released to the public with [7].”