This document summarizes different methods for monocular model-based 3D tracking of rigid objects using a survey format. It discusses edge-based methods like RAPiD that match model outlines to straight line segments extracted from images. It also reviews optical flow-based methods that track points between frames and techniques that combine optical flow with edge tracking. Additionally, it examines template matching approaches, interest point detection methods like Harris-Stephen and Shi-Tomasi, interest point matching using correlation windows, and pose estimation by tracking planes defined by sets of points. The document provides an overview of key algorithms for model-based 3D tracking and segmentation.
9. 4.1.2 Making RAPiD Robust Minimize the distance Control points lying on the same object edge are grouped into primitives. And a whole primitive can be rejected from the pose estimation. RANSAC methodology The number of edge strength maxima visible
10. 4.1.3 Explicit Edge Extraction The middle point, the orientation and the length of the segment Of a model segment Of a an extracted segment Mahalanobis distance Is the covariance matrix The pose is then estimated by minimizing
11. 4.2 Optical Flow-Based Methods Its corresponding location in the next image The projection of a point in an image at time
12. 4.2.1 Using Optical Flow Alone Normal optical flow For large motions Causes error accumulation
13. 4.2.2 Combining Optical Flow and Edges To avoid error accumulation Depends of the pose and the image spatial gradients at time Is a vector made of the temporal gradient at the chosen locations
14. 4.3 Template Matching To register a 2D template to an image under a family of deformations
15. 4.3.1 2D Tracking To find the parameters of some deformation That warps a template into the input image is the pseudo-inverse of the Jacobian matrix of computed at
16. 4.4 Interest Point-Based Methods Use localized features Rely on matching individual features across images and are therefore easy to robustify against partial occlusions or matching errors
17. 4.4.1 Interest Point Detection Harris-Stephen detector / Shi-Tomasi detector The pixels can be classified from the behavior of the eigen values of The coefficients of are the sums over a window of the first derivatives and of image intensities with respect to pixel coordinates
18. 4.4.2 Interest Point Matching to use7x7 correlation windows reject matches for which measure is less than 0.8 search of correspondents for a maximum movement of 50 pixels Kanade-Lucas-Tomasi tracker Keep the points that choose each other