文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 信息与通信 > Design and Implementation of DSP and FPGA-Based Robust
侵权投诉

Design and Implementation of DSP and FPGA-Based Robust

digital signal processor (DSP), FPGA implementation provides higher performance in repetitive and massive computations due to inherent architectural parallelism. Thus, in this paper, the position of the pendulum is measured with a machine vision system whose pipelined image processing is implemented on an FPGA device to meet real-time constraints. To enforce robustness to model uncertainty, and to attenuate disturbance and noise due to visual measurements, the mixed H2/H∞ control is applied to design the stabilizing controller.

The rest of this paper is organized as follows. Section 2 describes the relationship between the world coordinates and the image coordinates. The image processing algorithms for determining the displacement of the pendulum from the image data are introduced. Section 3 presents the hardware

implementation and architecture of the image processing algorithms on an FPGA device. In Section 4, the mathematical model of the inverted pendulum system is presented. The design of a mixed H2/H∞ controller is given in Section 5. The experimental results are given in Section 6. Finally, Section 7 contains some concluding remarks.

II. VISUAL MEASUREMENT AND IMAGE PROCESSING

ALGORITHMS

The intrinsic camera parameters are obtained using the camera calibration procedure proposed in [20]. λ is chosen to be the distance from the origin of the camera frame to the pendulum. The intrinsic camera parameters are given in TABLE I.

TABLE I

Intrinsic camera parameters

Parameter

Design and Implementation of DSP and FPGA-Based Robust

Value

λ 0.16 m fsx 20.3591×10 pixels/msyf -0.2207×10 pixels/mo -0.4186 pixels oy

Design and Implementation of DSP and FPGA-Based Robust

-0.2136 pixels

To determine the position of the marker on the image plane, the captured images are processed by a series of image processing algorithms. A sample of a raw image with a uniform background captured by the camera is shown in Fig. 1. The essential image processing algorithms [22] are described below.

The machine vision system is used to determine the angular displacement of the pendulum. A circular-shaped marker is A. Edge Detection placed on the top of the pendulum to simplify visual sensing Edges are the pixels at or around which a large change in and to increase the accuracy of visual measurements. gray-level intensity occurs. In most images, edges Determination of the position of the marker is based on the characterize object boundaries and are therefore useful for the perspective pinhole camera model [20, 21], which is given by segmentation [22] and identification of objects in a scene.

p K[R|t]P, (1) Edge detection is the process of locating edges in an image by

where p [x,y,1]T represents the two-dimensional detecting significant intensity changes. Most edge detection homogeneous coordinates of the image point in the image techniques are based on the idea of computing image

gradients. Let f (x, y) denote the grayscale level of the pixel at

coordinate system, P [X,Y,Z,1]Trepresents the

point (x, y) in the image coordinate system.

three-dimensional homogeneous coordinates of the object In this paper, the computation of the image gradient is point in the world coordinate system, and is a scalar factor. based on the Sobel operator [22], which approximates the The 3×3 matrix R and three-dimensional vector t are the partial derivatives by a mask operation. A general 3×3 mask extrinsic camera parameters that describe rotation and structure is defined as translation between the world frame and the camera frame,

respectively. K is the intrinsic camera parameter matrix given by fsxfs ox . K 0fosyy

The corresponding mask operation is given by 01 0 11

fx(x,y) f(x i,y j)wx(i,j) (2) Here, (ox, oy) is the principal point on the image coordinate

j 1i 1

system in pixels. fsx is the size of unit length in horizontal

11

pixels , fsy is the size of unit length in vertical pixels, and fsθ is

fy(x,y) f(x i,y j)wy(i,j) (3)

the skew of the pixels. The world frame is chosen to be j 1i 1coincident with the camera frame. Thus, the extrinsic camera

where wx(i,j) is the coefficient of the horizontal mask, and

parameters are

wy(i,j) is the coefficient of the vertical mask. Since the

1000

background of the scene is uniform and the objects in the [R|t] 0100 . scene are simple, the Sobel edge detector is adequate to

0010 extract the edges of the objects. Sobel edge detection was

Fig. 1. Image with a uniform background captured by the camera.

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email