Lesson 1 Computer Vision and OpenCV
Introduction
1. How robots “see” the world
For artificial intelligence, the ability to see is essential. And how robots see
the world involves machine vision, an important branch of artificial intelligence.
Machine vision is the idea that the robot takes human’s place to measure
and make judgments. The captured target will be converted into image signal
by image sensor, CMOS or CCD, and then the image signal will be transferred
to the specialized image processing system which will convert the image signal
into digitized signal according to the pixel distribution, brightness, color, etc.
Image system perform various operations on these signals to extract the
features of the target, so as to control the device in the field based on the
judgments.
Machine vision technology is commonly applied in intelligent transport
system and intelligent housing system.
2. Image Recognition Introduction
Image recognition is a crucial technique that uses computer to process
and analyze the image so as to recognize different targets.
Similar to human eyes, machine image recognition starts at the point
where there is huge variance or sudden change, and the features will
recognized one by one. Our brain controls our eyes to collect the major
features of the image and filter the redundant information, and then integrate
the major features into the complete visual image.
The process of computer image recognition is no different from that of
human image recognition, which is divided into four steps.
1) Acquire information: the light signal, sound signal, etc., are converted
1
into electric signal by the sensors to acquire the information
2) Image preprocessing: perform smoothing, denoising, etc., on the
image to highlight the major features of the image.
3) Feature extracting and selecting: extract and select the image features,
which is the pivotal step.
4) Image classification: make the recognition rules that is design classifier
based on the training result to get the main category of the features so as to
improve the recognition accuracy.
Image recognition is mostly applied in remote sensing image recognition
and robot vision.
3.OpenCV Introduction
OpenCV (Open Source Capture Vision) is a computer vision library for
free handling various tasks about image and video, for example display the
image collected by the camera and make the robot recognize the real object.
OpenCV is more eminent than PIL, the built-in image processing library in
Python. OpenCV provides complete Python interfaces, and Python3.5 and
opencv-python library file have been integrated in the provided image system.
4. How Images are Stored in Computer
How the images are stored in computer after they are recognized?
2
In general, a picture is composed of pixels and each pixel can be
represented by R, G and B components within 0-255. OpenCV stores each
pixel as a ternary array making it convenient to record all information of the
image. In addition, OpenCV records the data of three color channels of RGB
image in the order of BGR.
Besides, images of other standards (HSV) are stored as multivariate array.
An OpenCV image is a two-dimensional or three-dimensional array. An 8-bit
grayscale image (black-and-white images) is a two-dimensional array, and a
24-bit BGR image is a three-dimensional array.
For a BGR image, the first value of the element “image[0,0,0]” represents
Y--axis coordinate or the row number(0 represents the top). The second value
represents X-axis coordinate or column number (0 represents leftmost ). And
the third value represents the color channel.
Same as Python array, these array recording images can be accessed
individually to obtain the data of some color channel or capture a region of the
image.
