I.What is machine vision?
Machine vision is the vision of machines. In other words, it is to endow machines with visual perception, enabling them to have scene perception capabilities similar to those of biological vision systems.
Vision is our most powerful way of perception. We can obtain a lot of information about the surrounding environment through visual perception without actually touching it. After the advent of computers, people began to try to endow machines with the ability of visual perception.
Because the biological visual system is extremely complex, we are still unable to endow a machine system with such a powerful visual perception capability at present. At this stage, we are still committed to: building a machine vision system that can handle specific tasks in a controllable environment. As the visual environment in industry is controllable and the processing tasks are specific, most of the machine vision is currently applied in industry.
II. What are the main tasks of machine vision?
The main task of machine vision is to generate a set of descriptive information about the scene or object depicted in the image by analyzing the image.
That is to say, the input of the machine vision system is an image (or a sequence of images), and the output is the perception description of these images. These descriptions are closely related to the objects or scenes in these images, and these descriptions can help the machine complete specific subsequent tasks and guide the robot system to interact with the surrounding environment.
For example: Guide the robotic arm to grasp the parts on the conveyor belt as per the requirements. The types, positions and orientations of the parts are arbitrary. So when the parts on the conveyor belt pass by the camera above, through machine vision, a set of descriptions of the parts can be generated: type, position and orientation, thereby guiding the robotic arm to perform the grasping as required.
III. What is the relationship between machine vision and other related fields?
Machine vision is closely related to three fields: image processing, pattern classification, and scene analysis.
(1)Image processing mainly involves: generating a new image based on the existing one. Since the resulting image is still a single image, its output still requires human analysis and interpretation.)
(2) The main task of pattern classification is to categorize "patterns". These "patterns" refer to a set of attributes or characteristics of a thing. Based on these attributes and characteristics, they are classified into a certain category within the known classes, which means that the thing has been identified.
(3) The focus of scene analysis is to transform a simple description into a more complex, detailed one that is more conducive to our making judgments or drawing conclusions. These output descriptions are an elaboration of the input descriptions, and the output descriptions further explain the underlying connections of things.
So, I would like to reemphasize that the core issue that machine vision needs to address is: to obtain a symbolic description of an object based on a single image or a sequence of images!