CC BY
0
3. Hughes, Thomas, "The Finite Element Method - Linear Static and Dynamic Finite Element Analysis", Prentice Hall, Inc., (1987)
4. M. Geradin, D. Rixen, Mechanical Vibrations, Wiley, Chichester 1994.
5. M. West, C. Kane, J.E. Marsden, and M. Ortiz, "Variational integrators, the Newmark scheme, and dissipative systems", International Conference on Differential Equations, Berlin, 1999.
© Thuy B.T., 2024
Tran Thi Tram
Master at University of Mining and Geology,
Hanoi,Vietnam
DEVELOPING AN OBJECT IDENTIFICATION SYSTEM USING CAMERA FOR THE SYSTEM CLASSIFICATION OF PRODUCTS USING ROBOT
Annotation
The report summarizes the results of research conducted in the field of object recognition and positioning using cameras for robotic product classification systems. Machine vision is a potential direction for a variety of applications in the design and manufacture of devices for intelligent measurement and control systems.
Keywords:
robot, Machine vision
1. Introduction
Machine vision is a rapidly growing field with many applications in measurement and control (Peter I. Corke, 1996; Ramesh Jain, Rangachar Kasturi, Brian G. Schunck, 1995). Because vision sensor technology has made great strides in size and resolution, many smart devices with vision have appeared in medical technology, national security and defense, and the arts. entertainment. Many authors are interested in research on industrial image processing, especially the study of algorithms for identifying and locating objects in real time. In this research direction, I have built an object identification and positioning system used for product classification, a system for developing industrial image processing systems connected to robot control systems for the purpose of developing industrial image processing systems:
- Research and develop image processing systems for measurement and control.
- Research and develop object identification and positioning systems used for product classification systems.
- Building a camera image recognition and positioning system for industrial and defense applications.
- Building robot control algorithms based on image feedback.
- Develop dynamic image processing software for measurement and control applications.
Images are received by the camera, processed, identify objects, determine location, convert coordinates, using Labview software tool. The results of image recognition and positioning are passed to the robot control software module to control Robix to classify products.
The process of implementing recognition algorithms has been handled on two software, Imaq Vision Builder and Labview. The recognition part from the Camera will calculate the coordinates (position and direction angle) of the object and transfer those coordinates to the software that controls the robot handle to pick up the object.
HAyHHbIM WyPHAfl « IN SITU »
ISSN (p) 2411-7161 / ISSN (e) 2712-9500
№6 / 2024
2. Introduction to digital photography
Image information received through the Camera is processed and transmitted to the computer with characteristics such as image scanning time, resolution, and gray level of the image. Normally, the camera scans images and transmits them to the computer to display about 24-30 images/second through an image processing card. Each image is digitized and displayed on the screen as pixels. In fact, there are many different resolutions such as 256x256, 320x256, 480x640, 512x512. The gray levels of the image are also determined in many different ranges from 16 - 16383. If we have a 4-bit ADC, there will be 16 gray levels, 8 bits - 256 gray levels. The returned image frames are represented by the function f(x,y), the value of this function is the gray levels of the pixels at x,y coordinates on the screen. Images are saved to the computer for processing in the following standard formats: Bitmap Image (BMP) Tiff Image (TIFF) Png Photo (PNG) JPEG Photos (JPEG)
In this article, I used a camera connected via an image processing card and displayed on the computer at a speed of 24 images/second. Each received image has a resolution of 768x576 and uses 256 gray levels.
3. Image processing functions
After the image has been digitized and recorded in computer memory, we can use it to perform image processing tasks such as: Upgrading photos Disturbance handler Determine image characteristics Upgrading photos
When receiving a photo from the Camera, due to poor light contrast or too high glare or unclearness, we can use different methods to upgrade the photo, making it easier to see and handle. more reasonable. Common methods include histogram equalization, gamma corrections, logarithmic corrections and exponential corrections.
We will consider a method for adjusting histogram, this method is quite effective because it evenly stretches the gray level of the image within the allowable range. The function I(x,y) distributes the gray level intensity of each image pixel according to the x,y coordinate value. Use formula (1) to upgrade the image using the histogram correction method:
Imin , Imax - corresponding lowest and highest gray level value in the image I new, I old - the corresponding pixel gray level value of the new and old images
After performing the image enhancement process, we will receive an image with better contrast, facilitating the analysis and identification process.
After performing the image enhancement process, we will receive an image with better contrast, facilitating the analysis and identification process. Disturbance handler:
During the image acquisition process, noise handling is always a matter of concern. To handle the noise problem we use filters. Can be divided into 2 types of filters: Linear filter or called convolution Nonlinear filter
I
new
-1 ■ max ^ min
^min
256
(1)
The linear filtering algorithm will recalculate the value of a pixel based on the sum of the weights of neighboring pixels, and the kernel convolution coefficient. The size of the Convolution kernel is unlimited, you can use sizes 3x3, 5x5, 7x7, 127x127,...
Convolution can be divided into the following 4 types: Gradien, Laplacian, smoothing and Gaussian. Non-linear filters include: Gradient, Roberts, Sobel, Prewitt, Differentiation, Sigma, Median. In this noise processing problem, we used a Gaussian linear filter with a kernel size of 5x5, the calculation formula of the Gauss filter is:
(^ 2 + y 2) 2
g[ x, y] = e (2)
4. Find thresholds and identify image features
After the process of upgrading and filtering the image, to determine the image characteristics, we will find the image threshold to separate the object from the background. There are two commonly used threshold finding methods: Find the average threshold:
n n
Itb =ZZ y)/N2 (3)
i=1 j=1
i0,1(x, y) < I
1 (^y) Hi „ (4)
111 (X y) > Ing
Ing = Itb + 4- (5)
I - Threshold value
ng
N2 - Size of photo Find threshold according to Histogram:
The histogram is a graph that represents the number of pixels corresponding to each gray level of the image.
The algorithm for finding the threshold according to Histogram is determined as follows: Find 2 maximum points on the histogram line of the image, respectively Imaxi, Imax2 , the threshold value is determined:
t ^maxl + ^max 2
Ing =-2--(6)
5. Conclusion
The object identification and positioning system used for product classification has been built and operates reliably, allowing research and development of machine vision application systems. New methods of processing, filtering, identifying and locating images of objects are easily tested with rich software tools. References
1. Peter I. Corke (1996). Visual control of robots. Research Studies Press LTD.
2. Ramesh Jain, Rangachar Kasturi, Brian G. Schunck (1995). Machine Vision. McGraw-Hill. Inc.
©Tran Thi Tram, 2024
