Monday, May 26, 2014

New chessboard pattern

New chessboard pattern had been applied. The pattern is 8*5 in size, and each block is 26 mm.
The two sets (RGB and IR) of pictures used for calibration are shown as following:
Once images were taken, I applied them separately in GML calibration tool. It gives intrinsic matrices, extrinsic matrices, and distortion parameters.


The extrinsic matrices for RGB camera is:
42.92  73.12  316.15  -2.00  2.06  0.42
63.21  -48.93  319.51  -0.02  3.03  0.24
57.10  -36.59  366.05  -0.06  2.51  -0.29
39.41  -72.19  361.86  0.63  2.60  -0.16
-31.39  -55.33  219.10  -1.73  -2.10  -0.78
42.52  15.09  495.30  0.25  3.07  -0.03
47.00  -92.77  482.56  1.44  2.23  -0.59
The intrinsic matrices for RGB camera is:
524.10  0  312.14
0  524.91  249.84
0  0  1


The extrinsic matrices for IR camera is :
58.14  75.58  330.03  -1.99  2.04  0.40
80.93  -46.85  333.38  -0.03  3.00  0.26
72.56  -34.45  379.79  -0.05  2.48  -0.30
54.01  -68.20  381.87  0.63  2.56  -0.17
-12.85  -54.56  230.66  -1.74  -2.11  -0.80
54.45  16.40  518.28  0.25  3.02  0.02
57.72  -92.78  502.68  1.45  2.19  -0.57
The intrinsic matrices for RGB camera is:
612.32  0  326.70
0  614.54  242.53
0  0  1

Further calculation of mapping RGB to IR camera will be applied.

Saturday, May 24, 2014

Infra-red Image capturing


There are some existing applications such as Processing, MRPT which support IR Image capturing. However, they were all built for Xbox Kinect, and they does not support Kinect for Windows.
Furthermore, Kinect SDK does not provide a built-in IR image capture functionality, so I decided to use OpenCV, and OpenKinect (Libfreenect) to capture IR images.

OpenCV libraries are used for processing images. It is used in my application for creating/capturing frame from video, and saving the images.

OpenKinect libraries provide the drivers for Kinect in all system environment (Mac, Linux and Windows). And I used it for getting RGB video and IR video, then pass it to OpenCV to do further process.

Here are the images that I captured.(Left side was taken from RGB video, right side was taken from IR video):


The calibration tool that I'm going to use is GML calibration toolbox.

The chessboard pattern I was using does not fulfil the requirements of the tool, it requires odd * even in size.  New chessboard pattern will be applied.


Sunday, May 18, 2014

Kinect calibration Introduction

Kinect has two video cameras, one is for color image, another one is for Depth image.  And two sensors are located geometrical differently,therefore, the colour image and depth image are not pixel-to-pixel synced for a same scene. 


The sample data also contains a file of tracked points and its related Animation Unit. The tracked points data is recorded based on the coordinates of RGB video sensor, so in order to use these information for depth video, calibration is required. 

There are varieties existing tools for calibration, all I need to do is to capture several pictures of a chessboard pattern from both RGB and IR cameras.



Friday, May 16, 2014

Procedures of the project

As I discussed with my supervisor, the project can be done in about 5 steps, as following:

1. Kinect device calibration. 
2. Mapping RBG video to depth video or other way around.
3. LBP-TOP 
4. Feature extractions.
5. Classification.