OpenCV Tracking秩约束特征跟踪，改进了klt跟踪。 在恶劣的视觉条件下提供更好的跟踪.zip

Rank-Penalized Feature Tracking Author: Bryan Poling (Based on work with Gilad Lerman and Arthur Szlam) License: Modified BSD License (See License.txt) ************************************ Description ************************************ ******************************************************************************************* This code implements the rank-penalized feature tracker described in the paper "Better Feature Tracking Through Subspace Constraints", B. Poling and G. Lerman and A. Szlam, CVPR 2014. The suite is written in C++ and relies on the OpenCV computer vision library. The package consists of 3 programs: trackLive: Approximately real-time feature tracking using KLT, gradient descent, or the rank-penalized multi-feature tracker with either a centered or uncentered trackpoint matrix and with any of 3 supported dimension estimators (described in the above paper). Video is captured from a connected webcam (must be supported by OpenCV) and feature management is handled autonomously. Run trackLive -h for usage information. trackPostProcess: Feature tracking on pre-recorded data using KLT, gradient descent, or the rank-penalized multi-feature tracker of the associated paper. All options supported by trackLive are supported. The tracker can be run with several feature-management modes. Run trackPostProcess -h for usage details. degradeImage: The program we used for synthetically degrading high-quality source video for experimental comparison between different algorithms. Different cameras will have different characteristics, which may require adjusting tracker parameters. Multi-tracker parameters are set in the source code for the trackLive and trackPostProcess programs (trackLive.cpp and trackPostProcess.cpp, respectively). Also, some generic parameters (like feature size, number of image pyramid levels, and optimization settings) can be set in parameters.h. In this version of the code, changing parameters requires re-compilation of the project. ************************************** Building ************************************* ******************************************************************************************* This software was developed and tested on Debian Linux. In order for proper compilation, OpenCV must be properly installed, with C++ bindings (if installing from a repo make sure to install the -dev packages as well). OpenCV must be built with V4L2 support for trackLive to properly interface with a webcam under Linux. The Eigen linear algebra header library should also be installed. While it should be possible to build on non-*nix platforms, it will require setting up your own build process based on the make file included in this package (Windows will not support pkg-config, for instance), and Some camera feature (like managing things like autofocus) may not work on non-*nix platforms. There is nothing about the programs which should prevent one from porting to Windows or Apple OS, but it is not for the faint of heart. On Linux, with the proper dependencies, run "make" from the base directory to build all programs.