Presentation is loading. Please wait.

Presentation is loading. Please wait.

Session: Video Analysis and Action Recognition, Friday 9 November 2012

Similar presentations


Presentation on theme: "Session: Video Analysis and Action Recognition, Friday 9 November 2012"— Presentation transcript:

1 Session: Video Analysis and Action Recognition, Friday 9 November 2012
Click to edit title Click to edit text Sequential Reconstruction Segment-Wise Feature Track and Structure Updating Based on Parallax Paths Mauricio Hess-Flores1, Mark A. Duchaineau2, and Kenneth I. Joy1 1 Institute for Data Analysis and Visualization, University of California, Davis, USA 2 Lawrence Livermore National Laboratory, Livermore, CA, USA (Now at Google, Inc.) Good afternoon, I’m Mauricio and I will present the poster corresponding to the paper ‘Sequential Reconstruction Segment-Wise Feature Track and Structure Updating Based on Parallax Paths’. (7s) Session: Video Analysis and Action Recognition, Friday 9 November 2012

2 Parallax paths concept and constraints:
Sequential Reconstruction Segment-Wise Feature Track and Structure Updating Based on Parallax Paths Parallax paths concept and constraints: Parallax paths Intra-camera Inter-camera Motivation Concept Constraints Prediction (a) (b) Replicas (c) (d) Results: We present a novel method for multi-view sequential scene reconstruction scenarios such as in aerial video, that exploits the constraints imposed by the path of a moving camera, to allow for a new way of detecting and correcting inaccuracies in the feature tracking and structure computation processes, such as those as shown in (a). (15s) Our main contribution is to show that for short, planar segments of a continuous camera trajectory, parallax movement corresponding to a viewed scene point should ideally form a scaled and translated version of this trajectory when projected onto a parallel plane, depicted in (b). (12s) This creates both inter-camera and intra-camera constraints, as shown in (c), which allow for the detection of inaccurate feature tracks and a direct prediction of how to correct them to fit the consensus model from accurate feature tracks and given cameras, as shown in (d), leading to a simplified and more accurate scene structure computation. (15s) Results on real and synthetic aerial video and turntable sequences show that our method is able to correct outlier tracks, detect and correct tracking drift, and allow for an improvement of scene structure, additionally resulting in an improved convergence for bundle adjustment optimization. Please come see my poster and I will be happy to explain the algorithm in detail! Thank you. (15s) Total: 64s Feature track outlier detection and correction Drift detection and correction Improvement in scene structure Mauricio Hess-Flores1, Mark A. Duchaineau2, and Kenneth I. Joy1 1 Institute for Data Analysis and Visualization, University of California, Davis 2 Lawrence Livermore National Laboratory (now at Google, Inc.)


Download ppt "Session: Video Analysis and Action Recognition, Friday 9 November 2012"

Similar presentations


Ads by Google