Camera System for Movement Analysis

The goal of this work is to create an application for 3D motion analysis by using two cameras in the MATLAB software environment. Two synchronized video sequences are obtained from the two cameras with high definition (full HD) resolution, in which anatomically significant points have been labeled by blu markers. These video sequences are used for off-line processing by the means of computer vision, imag processing and pattern recognition. The application consists of a modular system, which contains color calibration, camera calibration, marker detection, stereo calibration, 3D reconstruction and motion analysis. Color calibration transforms (using multiple linear regression, linear and Hermite interpolating) the uncalibrated image to the RGB standard. Marker detection detects the coordinates of the markers using methods such as region prediction, multispectral thresholding, iterative thresholding, minimum error thresholding, Hough transform, unsupervised learning with Expectation-Maximization algorithm and K-means. Camera calibration, stereo calibration and 3D reconstruction result in obtaining the intrinsic and extrinsic camera parameters for the projection from 2D to 3D space. Motion analysis provides the basic characteristics of motion.

The goal of this work is to use MATLAB to create a functional application that can analyze 3D motion with the help of two cameras. Complete systems for 3D motion analysis, such as those produced by the firms Vicon and Qualisys, are expensive and use infrared cameras (up to 10) with high resolutions (up to 16 Megapixels) and reflexive markers. The camera system developed here uses two ordinary cameras with high definition (full HD) resolution. The video sequences from the cameras are used for off-line processing utilizing approaches of computer vison, image processing and pattern recognition. This system is sufficient for analyzing motion in clinical, and is specifically able to analyze hand motion of patients suffering from Parkinson’s disease.

Parkinson's disease (PD) is a progressive degenerative disorder of the central nervous system that often impairs the sufferer's motor skills and causes specific disturbances of hand motor function. Tremors, rigidity, poor balance and difficulty walking are characteristic primary symptoms of PD. Lesions in different structures of the central and peripheral nervous systems cause specific disturbances of hand function in the resting position (e.g. resting tremor and dystonia in PD), when movement is initiated (PD), in reaching a target (cerebellar disturbance), etc. It can be assumed that different lesions would also influence the various phases of a manual transport movement such as: forming a grip on an object, establishing the grip, lifting the object, the transport phases, and placing the object on the target point. A method based on a simple manual transport act could therefore be useful for an objective description and quantification of certain hand movement disturbances. The successive steps for motion analysis are suggested in the flowchart in Fig. 1.2. and the proposed modular camera system is shown in Fig. 1.3:. The parts of the camera system are described in the following chapters. In the next chapter the state of the art of similar camera systems and their approaches are discussed. In chapter 3 image preprocessing consisting of deinterlacing methods, motion blurring and color calibration is introduced. Basic deinterlacing methods for preprocessing videos and use of deinterlacing methods as well as results are discussed in regards to the camera system. Color calibration, including used methods, experiments and results, is described in its own subchapter. The parts of fundamental theory and results for the camera calibration, stereo calibration, and 3D reconstruction are mentioned in chapters 4 and 6 with more references for future study. Marker detection is introduced in chapter 5, which describes the methods used for marker segmentation. Many algorithms based on different approaches are tested and subjected to experiment. Algorithms such as multispectral thresholding, Hough transform, Expectation-Maximization (EM) algorithm and K-means are compared. K-means algorithm is described as a relation to the EM algorithm. Other techniques are discussed, which can be used for applications such as iterative thresholding, minimum error thresholding based on the Gaussian approximation of an image histogram and so on. In chapter 7 basic characterization of motion is discussed. The conclusion in chapter 8 contains a summarization of the implemented algorithms, achieved experiments and results. In the appendix it is possible to see some interesting results considered by the author (unsupervised learning process with EM algorithm, some results of color calibration, global Hough transform for circles, evaluated normal, circle and jerky motion and so on).