StereoScan: Dense 3d reconstruction in real-time

By Andreas Geiger, Julius Ziegler, Christoph Stiller. 2011 IEEE Intelligent Vehicles Symposium (IV), 2011

1,184 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/8e7b64024ed4c5cf233c7d678234eb8328b0b9d5

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

Joint 3d Estimation of Vehicles and Scene Flow (2015, methodology): “As precomputed disparity estimates (Hirschm¨uller, 2008) and optical ﬂow features (Geiger et al., 2011) are not available for every pixel, we calculate C sparsex only at locations for which observations exist.”
D Traffic Scene Understanding from Movable Platforms (2013, methodology): “We compute the vehicle’s egomotion and 3D scene flow using the method presented in [34].”
Omnidirectional 3D reconstruction in augmented Manhattan worlds (2014, methodology): “feature points in the previous frame t − 1 and estimate motion by minimizing the reprojection error with respect to the observations in the current frame t similar to the StereoScan system [40] for perspective cameras.”
DeepVO: Towards end-to-end visual odometry with deep Recurrent Convolutional Neural Networks (2017, methodology): “1, employ multi-view geometry [5] to determine motion after extracting and matching (or tracking) salient feature points from a sequence of images, such as the algorithm in [6] and LIBVISO2 [7].”
Efficient sparse-to-dense optical flow estimation using a learned basis and layers (2015, methodology): “Then, we use the features from [17], which are designed for visual odometry applications.”
PROBE-GK: Predictive robust estimation using generalized kernels (2016, methodology): “We used the open-source library LIBVISO2 [10] for feature extraction and matching.”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, methodology): “Geiger et al. (2011) present a real-time 3D reconstruction approach using visual odometry.”
Efficient Optical Flow and Stereo Vision for Velocity Estimation and Obstacle Avoidance on an Autonomous Pocket Drone (2016, background): “Since there are strict limitations on energy expenditure, sensing, and processing capabilities on a pocket drone, even relatively efficient stereo vision methods [2], [3] are computationally too heavy to run on-board a microprocessor.”
1 year, 1000 km: The Oxford RobotCar dataset (2017, methodology): “These datasets focus primarily on the development of algorithmic competencies for autonomous driving: motion estimation as in Geiger et al. (2011) and Nistér et al. (2006), stereo reconstruction as in Geiger et al. (2010) and Hirschmuller (2005), pedestrian and vehicle detection as in Benenson et…”
Nonparametric semantic segmentation for 3D street scenes (2013, methodology): “Geometric accuracy evaluation: By following [24], we using laser measures from Velodyne to evaluate the 3D model.”
Are we ready for autonomous driving? The KITTI vision benchmark suite (2012, neutral): “We evaluate ﬁve different approaches on our visual odometry / SLAM dataset: VISO2-S/M [21], a real-time stereo/monocular visual odometry library based on incremental motion estimates, the approach of [1] with and without Local Bundle Adjustment (LBA) [32] as well as the ﬂow separation approach of [2”
PROUD—Public Road Urban Driverless-Car Test (2015, methodology): “The VOD is composed by a stereo-based 3D world reconstruction, an obstacle clustering which uses a occupancy grid map, and a tracking stage [25].”
Generic 3D Representation via Pose Estimation and Matching (2016, methodology): “…We compared against Structure-from-Motion (visualSfM [59,60] with default components and tuned hyper-parameters for pair-wise pose estimation on 192 × 192 patches and full images) and LIBVISO2 Visual Simo - Serra ( pre-trained ) m= 2.0 ×10 -3 F r e qu e n c y ( % ) Odometery [21] on full images.”
3D Traffic Scene Understanding From Movable Platforms (2014, methodology): “We compute the vehicle’s egomotion and 3D scene flow using the method presented in [34].”
Multi-scale object candidates for generic object tracking in street scenes (2016, methodology): “First, we transform the 3D object proposals to the common coordinate frame using visual odometry of [25] and link them across”
Probabilistic data association for semantic SLAM (2017, methodology): “Rather than use inertial odometry in this experiment, we used the VISO2 [46] visual odometry algorithm as the initial guess X (0) for a new keyframe state.”
Visual Semantic Search: Retrieving Videos via Complex Textual Queries (2014, methodology): “We use the Hamming loss as the loss function, which is decomposable as ∆(y,y(i)) = ∑ uv 1(yuv 6= y(i)uv ) = a(i) − ∑ uv yuvy (i) uv , where a(i) , ∑ u c (i) u is the total number of matching edges, which is a constant.”
Visual SLAM and Structure from Motion in Dynamic Environments (2018, methodology): “It turns out that by utilizing both CNN and RNN, the output odometry is much better and has competitive performance over the state-of-the-art methods (compared to VISO2 Monocular and Stereo system [45]).”
BA-Net: Dense Bundle Adjustment Network (2018, methodology): “Since ground truth pose is unavailable from the raw KITTI dataset, we compute camera poses by LibVISO2 (Geiger et al., 2011) and take them as ground truth after discarding poses with large errors.”