Self-supervised Monocular Road Detection in Desert Terrain

By Hendrik Dahlkamp, Adrian Kaehler, David Stavens, Sebastian Thrun, Gary R. Bradski. 2006

405 citations recorded by the source corpus.

We present a method for identifying drivable surfaces in difficult unpaved and offroad terrain conditions as encountered in the DARPA Grand Challenge robot race. Instead of relying on a static, pre-computed road appearance model, this method adjusts its model to changing environments. It achieves robustness by combining sensor information from a laser range finder, a pose estimation system and a color camera. Using the first two modalities, the system first identifies a nearby patch of drivable surface. Computer Vision then takes this patch and uses it to construct appearance models to find drivable surface outward into the far range. This information is put into a drivability map for the vehicle path planner. In addition to evaluating the method’s performance using a scoring framework run on real-world data, the system was entered, and won, the 2005 DARPA Grand Challenge. Post-race log-file analysis proved that without the Computer Vision algorithm, the vehicle would not have driven fast enough to win.

Subfields: Aerospace Engineering, Automotive 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 Automotive Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation (2015, background): “image search [1], robotics [2], medical imaging [3], object detection in radar images [4] or face detection [5].”
Road Detection Based on Illuminant Invariance (2011, methodology): “Some authors propose the use of depth as an additional cue for adding robustness, as in [ 9 ], in which depth comes from active sensors, or in [10], in which depth comes from as tereo rig.”
Deep belief net learning in a long-range vision system for autonomous off-road driving (2008, background): “Although this approach helped to win the 2005 DARPA Grand Challenge, its utility is limited by the inherent fragility of color-based methods.”
Towards Real-Time Object Detection on Embedded Systems (2018, background): “detecting the locations and categories of multiple object instances in one single image, is an essential problem in a large variety of applications such as robotics [9] and self-driving vehicles [10].”
Individual Tree-Crown Detection in RGB Imagery Using Semi-Supervised Deep Learning Neural Networks (2019, background): “One approach to addressing the data limitation in deep learning is “self-supervised learning” (sensus [14]), which uses unsupervised methods to generate training data that is used to train supervised models [15].”
D Traffic Scene Understanding from Movable Platforms (2013, neutral): “net/projects/intersection [18], [21], [41], [49] and road detection [1], [2], [17] algorithms.”
Tracking-based semi-supervised learning (2011, background): “For example, Dahlkamp et al. (2006) trained a vision-based road detection system for autonomous driving on labels provided by a laser range finder, significantly extending the range of road detection while requiring no extra human supervision.”
Incremental dense semantic stereo fusion for large-scale semantic scene reconstruction (2015, background): “1, is a crucial prerequisite for a variety of applications, including robot navigation [1], [2], semantic mapping [3], [4], wearable and/or assistive technology [5], [6], and change detection [7].”
3D Traffic Scene Understanding From Movable Platforms (2014, background): “This is mainly due to the complexity of urban trafﬁc situations and the limited capabilities of current onboard sensors and image understanding algorithms.”
Finding multiple lanes in urban road networks with vision and lidar (2009, neutral): “Other environmental cues such as color and texture may be useful in such situations (Dahlkamp et al. 2006).”
Learning long‐range vision for autonomous off‐road driving (2009, background): “This is especially useful for roadfollowing vehicles (Dahlkamp et al., 2006; Leib et al., 2005; Hong et al., 2002); the ground immediately in front of the vehicle is assumed to be traversable, and the rest of the image is then filtered to find similarly colored or textured pixels.”
Guaranteed Safe Online Learning via Reachability: tracking a ground target using a quadrotor (2012, background): “Their use has enabled robots to perform tasks that were previously impossible, from ﬂying complex aerobatic maneuvers using an autonomous helicopter [1], to quickly traversing challenging terrain using a quadruped [2], to navigating hundreds of miles of desert roads using an autonomous car [3].”
Vision-Based Offline-Online Perception Paradigm for Autonomous Driving (2015, background): “Autonomous navigation in urban environments requires a deep and fast understanding of the scene for safe decision making regarding path planning and obstacle avoidance [2, 12, 33].”
Automatic dense visual semantic mapping from street-level imagery (2012, background): “tasks such as determining drivable regions for autonomous vehicles [7], navigating structured environments [25], and interacting with objects [5].”