The 9th International Conference on Ubiquitous Robots and Ambient Intelligence (URAI 2012) Nov. 26-28, 2012 in Daejeon Convention Center(DCC), Daejeon, Korea

Milk: Monocular Visual Odometry with Motion Constraints Sunglok Choi and Wonpil Yu Robot and Cognitive Convergence Research Dept., ETRI, Republic of Korea (E-mail: [email protected], Web: http://sites.google.com/site/sunglok)

Keywords - Monocular Visual Odometry, Planar Motion, Circular Motion, 1-point RANSAC Visual odometry is a task to estimate ego-motion of a camera using its stream of images. Recently it has gained more attention because its applications to outdoor autonomous navigation. When autonomous vehicles are operated on rough terrain, they undergo 6-DoF three-dimensional motion which cannot be estimated by wheel odometry. Wheel odometry also suffers from error by wheel slippage. Sometimes autonomous cars are difficult to use wheel odometry because rotary encoders should be placed directly on wheels or their shafts. Many researches on visual odometry are based on stereo cameras because they provide image depth additionally. Such depth information enables to estimate camera trajectory with proper scale. In contrast, visual odometry with a single camera cannot estimate length of translational motion due to scale ambiguity. To resolve scale ambiguity, it needs additional information such as known initial motion or structures. However, a single camera has still advantages in the point of cost and simplicity of hardware. Motion Constraints Our monocular visual odometry is based on additional motion constraints, planar and circular motion, proposed by Scaramuzza et al. [1]. Many indoor and on-road vehicles exhibit 3-DoF planar motion. The epipolar constraint under planar motion has two unknowns, θ and φ , where θ is angle of rotation, φ is orientation of translation. We can estimate two unknowns using at least two pairs of features (a.k.a. 2-point algorithm) [2]. Even though motion is not completely circular, its small segments can approximate to circular motion with various curvatures. Circular motion leads additional constraints, θ = 2φ , so at least a pair of features is enough to find θ or φ (a.k.a. 1-point algorithm) [2]. Additionally we also develop 3-point algorithm to take into account pitch motion due to bumps and ramps. Monocular Visual Odometry Our visual odometry is composed of four steps: feature extraction/matching, outlier rejection, motion estimation, and motion accumulation. Feature extraction and matching associate features between adjacent images. We utilize optical flow imThis work was supported partly by the R&D program of MKE and KIAT. (Project: 3D Perception and Robot Navigation Technology for Unstructured Environments, M002300090)

Ground Truth VISO2−M Milk

150 100 50 Z [m]

Abstract - This video introduces about monocular visual odometry based on planar and circular motion assumption. Many indoor and on-road vehicles show planar motion, and each small piece of motion can approximate to circular motion. We presents that the motion constraints generate more accurate trajectories more quickly.

0

−50 −100 −150 −300

−200

−100 X [m]

0

100

Fig. 1: Two algorithms were performed on KITTI-07 dataset [5] with its ground truth. VISO2-M is monocular version of libviso2 [3], and Milk is the demonstrated visual odometry with motion constraints. plemented in libvio2 [3]. Outlier rejection is inevitable because matched features cannot agree with camera motion when they are wrongly matched or extracted from moving objects. We adopt 1-point RANSAC [1] which uses 1-point algorithm and a geometric error function by Choi et al. [4]. Since 1-point algorithm uses only one pair of features under the motion assumption, RANSAC needs significantly less number of iterations for outlier removal. We can estimate relative pose between adjacent images using only features corresponding to camera motion. We use 1-point, 2-point, 3-point algorithms, which were selected by the predefined firewall threshold [1]. To resolve scale ambiguity, we assume that the interval between the ground and camera is constant. Scale of motion simply results from the ratio of the interval in the real world and its value in the reconstructed structure. Finally, we can get camera trajectory by accumulating scaled relative pose.

References [1] D. Scaramuzza et al., “Real-time monocular visual odometry for on-road vehicles with 1-point RANSAC,” in ICRA, 2009. [2] S. Choi et al., “Numerical solutions to relative pose problem under planar motion,” in URAI, 2010. [3] A. Geiger et al., “StereoScan: Dense 3D reconstruction in real-time,” in IV, 2011. [4] S. Choi et al., “What does ground tell us? monocular visual odometry under planar motion constraint,” in ICCAS, 2011. [5] A. Geiger et al., “Are we ready for autonomous driving? the KITTI Vision Benchmark Suite,” in CVPR, 2012.

Milk: Monocular Visual Odometry with Motion Constraints

Abstract - This video introduces about monocular visual odometry based on planar and circular motion assump- tion. Many indoor and on-road vehicles show planar mo- tion, and each small piece of motion can approximate to circular motion. We presents that the motion constraints generate more accurate trajectories more ...

83KB Sizes 0 Downloads 174 Views

Recommend Documents

On Combining Visual SLAM and Visual Odometry - The University of ...
monocular SLAM system which combines the benefits of these two techniques. ... that end we recast the usual world-centric EKF implementation of visual SLAM ...

Face Tracking and Recognition with Visual Constraints in Real-World ...
... constrain term can be found at http://seqam.rutgers.edu/projects/motion/face/face.html. ..... [14] Y. Li, H. Ai, T. Yamashita, S. Lao, and M. Kawade. Tracking in.

Robust Direct Visual Odometry using Mutual Information
Differences based Lucas-Kanade tracking formulation. Further, we propose a novel approach that combines the robustness benefits of information-based measures and the speed of tra- ditional intensity based Lucas-Kanade tracking for robust state estima

Online Visual Motion Estimation using FastSLAM with ...
Estimating vehicle motion online is critical to creating truly autonomous systems. ... 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems ...

Resolving Scale Ambiguity for Monocular Visual ...
times help to overcome the scale ambiguity. ... public datasets: the KITTI odometry dataset (on-road) ..... “Real-time monocular visual odometry for on-road.

Boosting with pairwise constraints
Jul 16, 2009 - Department of Automation, Tsinghua University, Beijing 100084, China. Abstract ...... straints that lead to the same surrogate function. Since we ...

Inducing Herding with Capacity Constraints
This paper shows that a firm may benefit from restricting capacity so as to trig- ger herding behaviour from consumers, in situations where such behavior is ...

Contractual Pricing with Incentive Constraints
integral part of a team's organization when individual behavior is subject to incen- tive compatibility. (Without incentive compatibility, there is no need for secrets.).

Learning with convex constraints
Unfortunately, the curse of dimensionality, especially in presence of many tasks, makes many complex real-world problems still hard to face. A possi- ble direction to attach those ..... S.: Calculus of Variations. Dover publications, Inc (1963). 5. G

String Constraints with Concatenation and Transducers Solved ...
path-queries for graph databases [Barceló et al. 2013; Barceló et al. 2012], which has ...... important for our purpose. However, the crucial point is that all queries that a DPLL(T) solver asks ...... In USENIX Security Symposium. http://static.us

Modeling Preferences with Availability Constraints
it focuses our attempt of prediction on the set of unavailable items, using ... For instance, a cable TV bundle is unlikely to contain all the channels that ... work in this area in two ways. First, in ... [8], music [9], Internet radio [10] and so o

Kinect in Motion - Audio and Visual Tracking by Example ...
Kinect in Motion - Audio and Visual Tracking by Example - , Fascinari Massimo.pdf. Kinect in Motion - Audio and Visual Tracking by Example - , Fascinari Massimo.pdf. Open. Extract. Open with. Sign In. Main menu. Displaying Kinect in Motion - Audio an

Integration of visual motion and locomotion in ... - Matteo Carandini
Nov 3, 2013 - separate data segment (the 'test set'), we defined a prediction quality ..... learn the stable mapping between movements and visual flow. In a.

Statistical Constraints
2Insight Centre for Data Analytics, University College Cork, Ireland. 3Institute of Population Studies, Hacettepe University, Turkey. 21st European Conference on ...

Integration of visual motion and locomotion in ... - Matteo Carandini
Nov 3, 2013 - activity by 50% to 200% (Supplementary Fig. 6). ..... Previous data suggested that the effect of locomotion was binary6, as would be expected if ...

Motion integration and postdiction in visual ... - Semantic Scholar
176, 343 (1977); R. Wetts, G. N. ... A series of psychophysical experiments yields data inconsistent ... 17 MARCH 2000 VOL 287 SCIENCE www.sciencemag.org.

Monocular Obstacle Detection
Ankur Kumar and Ashraf Mansur are students of Robot Learning Course at Cornell University, Ithaca, NY 14850. {ak364, aam243} ... for two labeled training datasets. This is used to train the. SVM. For the test dataset, features are ..... Graphics, 6(2

visual motion estimation and terrain modeling for ...
Sep 8, 2005 - stereo cameras to localize a rover offers a more robust solution ... vision-based localization system to allow a planetary ..... The final bit file was.

Compression release retarder with valve motion modifier
Dec 8, 1988 - A process and apparatus are provided to increase the retarding horsepower .... With compression ignition engines having a fuel in jector driven ...

Compression release retarder with valve motion modifier
Dec 8, 1988 - and the master piston driven by a remote exhaust valve pushtube or the fuel .... top dead center position of the engine piston so as to deliver rapidly a ... of the retarding cycle (about 140 crankangle degrees). FIG. 10 is a ...

Aavin Milk-Vellore Cooperative Milk Producer's Federation Limited ...
Aavin Milk-Vellore Cooperative Milk Producer's Federa ... Executive and Various Post Application Form 2016.pdf. Aavin Milk-Vellore Cooperative Milk Producer's ...

heat-treated milk products fermented with Bacteroides ... - DTU Orbit
Jul 23, 2017 - used for the fermentation process with B. xylanisolvens DSM 23964. ... the microbiological data provided do not give rise to safety concerns. .... highest in two- to five-year-old children, with mean and 90th percentile values of 2 ...

Milk trait heritability and correlation with heterozygosity ...
yak was raised mainly by the Kham Tibetan and the plateau-type by the Amdo Ti- betan in China (WU 1999). About 15 ... The Jiulong yak is the typical valley-type raised by the Kham Tibetan people for many centuries in western ... using a controlled br