Machine vision inspection is widely adopted in industry because it maximizes productivity and reduces costs. Furthermore, machine vision systems can operate 24/7 with 100% online inspection at high speeds and near-100% accuracy. Machine vision is a relatively new technology that utilizes opto-mechatronics to endow machines with visual capabilities. Introducing machine vision into the inspection field enables high-precision, high-speed online measurements in many situations. The theoretical framework of machine vision inspection technology has also developed significantly, and its development can be divided into several stages and trends.
1. Primary Vision Theory: Primarily addressing the inverse problem of optical imaging, it consists of a series of processes that recover the physical properties of a three-dimensional visible surface from a two-dimensional light intensity array. The input data and computational objectives of each process are clearly describable, such as edge detection, stereo matching, and motion-based structure recovery. During the projection of a three-dimensional object into a two-dimensional image, significant amounts of three-dimensional information are lost, leading to ill-posed problems. Therefore, strengthening research on primary vision processes and their constraints is crucial, primarily focusing on 3D reconstruction and binocular vision.
2. Active Vision Theory: Active vision refers to the technical methods by which an observer moves in a definite or indefinite manner to track targets and perceive objects. In active vision, the observer and the target object can move simultaneously. The observer's movement provides additional conditions for studying the shape, distance, and motion of the target. Important research directions include target tracking and missile interception.
3. Visual information fusion: The fusion of multiple visual information sources may overcome the limitations of acquiring single visual information, and achieve the goal of acquiring static and instantaneous visual information in an ideal environment, thus meeting the requirements for understanding the complex objective world. The main research area is image information fusion.
4. 3D Scene Reconstruction: Current theories and algorithms for 3D scene reconstruction are limited to the "visible" parts of the scene, representing 2.5D information and providing only 3D information within the visible outlines of objects. Reconstructing complete information of both visible and invisible parts of the scene surface is a complex but urgent theoretical problem that needs to be solved.
5. Algorithm Performance Evaluation: Machine vision research focuses on whether a task is feasible or achievable, lacking the characterization and evaluation of the performance quality of algorithms and systems. In practical applications, efficiency and performance are crucial; otherwise, algorithms and systems cannot leave the laboratory. Therefore, establishing a performance evaluation system for machine vision algorithms is essential.
6. Visual Parallel Computing: Real-time visual computing still faces many theoretical, algorithmic, and technical challenges. The development trend of visual parallel computing architectures is to use increasingly smaller processing units within increasingly larger structures, with the future direction being the formation of massive processing network systems composed of basic logic operation processing units.
7. General-purpose visual information system: A general-purpose visual information system capable of performing various visual tasks. That is, to establish a machine vision system that is analogous to the function of the human visual system. Through the establishment of a dedicated visual system platform, it gradually develops into a complete general-purpose visual system, such as a visual platform and a highly intelligent visual robot.
