High-speed vision examples utilizing OpenCV

This project holds supplementary source files for the following tutorial article:

• Shingo Kagami: Utilizing OpenCV for High-Speed Vision Processing, Journal of the Robotics Society of Japan, 31(3), pp. 244-248, 2013 (in Japanese).

The objective is to present a sample implementation of high frame rate vision processing with OpenCV, in which a separate GUI thread is arranged so that it does not disturb a higher-rate vision processing thread as best it can. The implementation runs a target tracking demonstration at over 450 fps with a Point Grey Research USB 3.0 camera and a note PC.

OVERVIEW

Basically, the two (or more) threads run independently while the communication between them is done in a message-passing manner. The message passing link is designed with the policies and assumptions that (1) overhead of data copy is minimized, (2) the slower GUI thread should not disturb the faster vision processing thread, and (3) a message may be dropped if the slower receiving thread cannot handle it.

The project contains three modules:

• msglink: a simple thread-to-thread message-passing utility designed with the above policies.

• multi_threaded_cap: a sample program using msglink that simply captures and displays a video sequence from a standard web camera (which OpenCV can recognize).

• target_track: a sample program using msglink that demonstrates a template-match tracking with a Point Grey Research camera (which the Point Grey FlyCapture driver can recognize).

DEPENDENCY

• msglink depends on Boost (tested with Boost 1.47.0)

• multi_threaded_cap depends on OpenCV 2.x (tested with OpenCV 2.3.1), Boost and msglink.

• target_track depends on OpenCV, Boost, FlyCapture (tested with FlyCapture2-2.3.3.19) and msglink.

They are tested in Windows 7 Professional (x64) with Visual Studio 2008, but are not meant to be dependent on the Windows platform.

BUILD AND RUN

Make sure the above prerequisite libraries are installed and the include/library paths are appropriately set. Note that we are using the old FlyCapture 1.x API and thus FC1 directory of the FlyCapture library needs to be in the library path.

If you are using Visual Studio, start highspeed_cam_opencv.sln to open the solution. The projects for multi_threaded_cap and target_track are in it, for which the following settings are already done.

multi_threaded_cap:

• Link with opencv_core***.lib, opencv_imgproc***.lib, and opencv_highgui***.lib (where *** are the version number).

• Put the path where msglink.hpp lies into the additional include directory (e.g. ../msglink).

• When run, it requires a camera recognizable by OpenCV.

• Hit any key to quit.

target_track:

• Link with opencv_core***.lib, opencv_imgproc***.lib, opencv_highgui***.lib (where *** are the version number), and PGRFlyCapture.lib.

• Put the path where msglink.hpp lies into the additional include directories setting (e.g. ../msglink).

• When run, it requires a monochrome Point Grey Research camera recognizable by the FlyCapture driver. To use another camera (say, a color camera), see the comment around the class VideoCaptureFlyCap in target_track.cpp.

• Left click to choose (the center of) a new target.

• Hit any key to quit.