Computer Vision is a field of artificial intelligence that enables computers to analyse visual data—such as photographs, video, camera feeds and other images of spatial reality—in order to identify and interpret objects, scenes and processes.
In some forms, computer vision seeks to reproduce or approximate particular capabilities of human vision: recognising objects, distinguishing shapes, locating things in space, estimating depth and movement, following events and using visual information to guide actions. From the perspective viewpoint, these tasks depend heavily upon analysing perspective views and images of spatial reality.
Computer vision therefore provides an important modern connection between perspective, artificial intelligence, optics, imaging, measurement and three-dimensional spatial understanding. A camera captures a projected image of the world; the computer then attempts to extract useful information and knowledge about the objects, relationships and processes represented within that image.
Robotic vision is a particularly important subtype of computer vision concerned with live visual sensing, spatial interpretation and associated real-time decision-making by robots, autonomous vehicles and other artificial systems.

What Is Computer Vision?
Computer vision attempts to obtain useful information from visual images and image sequences. Instead of treating a photograph or video simply as something to be looked at, the computer analyses the image as data.
Depending upon its purpose, a computer-vision system may attempt to determine:
- what objects or processes are present;
- where they are located;
- their apparent size and shape;
- their orientation or direction;
- their relationship to other objects;
- their depth or distance;
- whether they are moving;
- how their appearance changes over time;
- whether two images contain the same object or scene;
- how several views relate to one another; and
- what action should be taken in response to the observed situation.
The problem is exceptionally complex because a two-dimensional perspective image is not a complete copy of the three-dimensional spatial reality that produced it. Some information is preserved, some transformed, some hidden and some lost.
Computer Vision and Perspective
Perspective is fundamental to computer vision because most ordinary camera images are projected views of three-dimensional spatial reality.
The original spatial scene exists in object or target space. A camera and optical system transform aspects of that scene into an image or perspective space. Computer vision then attempts to analyse that image and recover useful information about the original object, scene or event.
This relationship can be expressed simply as:
Spatial Reality → Camera / Imaging Process → Perspective Image → Computer Analysis → Information / Knowledge / Action.
Computer vision therefore occupies an important position within the larger perspective image chain. It does not merely form images; it attempts to decode, measure, classify and interpret them.
Goals and Functions of Computer Vision
The principal goals of computer vision closely overlap with several fundamental functions of perspective.
A perspective image or view can provide information about object or scene location, size, shape, scale, position, orientation, arrangement, texture, colour, state and activity. Computer vision attempts to extract, compare and use such information systematically.
Three especially important perspective functions are:
- Viewing — observing and capturing views or images of spatial reality.
- Matching — measuring, comparing, identifying, classifying, calculating, orienting and cross-matching visual information.
- Representing — modelling, mapping, indexing, linking and reconstructing aspects of spatial reality.
These functions are directly relevant to artificial vision. A computer may first capture or receive an image, then identify and measure its contents, compare them with other information and finally construct a useful classification, spatial model or action.
Perspective Products and Visual Information
Computer vision can also be understood in terms of the products of perspective: the kinds of information obtained from an image or view.
These can include information about:
- whole and part — identifying an object and its component structure;
- figure and form — recognising shape, contour and configuration;
- size and scale — estimating apparent or physical dimensions;
- position — locating objects within image or spatial coordinates;
- state — determining the condition of an observed object or process; and
- activity — interpreting change, movement or behaviour.
The importance of computer vision is therefore not merely that a machine can receive an image, but that it can potentially turn visual information into structured knowledge about spatial reality.
Two Broad Operating Contexts
For present purposes, two broad operating contexts of computer vision can be distinguished.
1. Live or Real-Time Computer Vision
Live computer vision analyses images as they are being captured so that an artificial or human-controlled system can respond to current events.
Examples include autonomous vehicles, robots, security systems, manufacturing equipment and traffic-control systems.
Here, visual analysis becomes part of a continuous cycle:
Observe → Analyse → Decide → Act → Observe Again.
2. Non-Live or Database Image Analysis
Computer vision can also analyse previously captured photographs, videos, scans and image databases.
Such analysis may classify images, identify objects, compare visual records, measure changes, search large image collections or extract statistical and spatial information.
The two operating contexts share many methods and goals, but real-time systems additionally face the problem of making useful decisions quickly enough to affect a changing physical situation.
Image Classification
Image classification attempts to determine the category or class to which an image, object or visual pattern belongs.
A system may distinguish, for example, between different kinds of objects, structures, scenes or events. In perspective terms, classification is closely related to the function of matching: visual information is compared with learned, stored or otherwise established classes.
Classification can operate on a complete image or on individual objects or regions contained within it.
Object Detection and Identification
Object detection concerns finding objects within an image or video, while object identification or recognition concerns determining what those objects are.
This distinction is important. A machine may determine that an image contains several separate objects before identifying them individually as a person, vehicle, building, tool, animal or other class.
The apparent form of the same physical object can change dramatically with viewpoint, distance, orientation, lighting, occlusion and optical conditions. Object recognition therefore confronts a fundamental perspective problem: how can one physical object be recognised across many different perspective appearances?
Shape and Form Recognition
Shape is one of the principal sources of information available to both human and computer vision.
However, the visible shape of a three-dimensional object is not fixed. Its projected form changes as the viewing direction changes. A circular surface can appear elliptical; a rectangular surface can become a general quadrilateral; an object can become strongly foreshortened or partially concealed.
Computer vision must therefore distinguish between physical form and perspective appearance. Recognising an object cannot depend simply upon matching one fixed outline with another.
This problem connects computer vision directly with perspective phenomena such as aspect, foreshortening, diminution, occlusion and degradation of form.
Edges, Contours and Features
Computer vision can analyse edges: significant local changes in image brightness, colour, texture, depth or surface orientation.
Edges may correspond to object boundaries, surface markings, shadows, texture changes or image noise. They can therefore provide useful information for object recognition, face recognition, image analysis and other processes.
An occluding contour is particularly important to perspective because it marks the visible boundary between a projected object surface and the surrounding scene.
Edges and contours are valuable sources of visual information, but they are not sufficient by themselves to describe the complete three-dimensional form of an object.
Scene Analysis
Computer vision frequently needs to understand not only individual objects but the structure of an entire scene.
A spatial scene may contain many interacting perspective relationships involving:
- objects and backgrounds;
- whole and part relationships;
- foreground, middle-distance and background;
- overlapping contours;
- relative size;
- texture gradients;
- shadows and lighting;
- visible and hidden surfaces;
- movement;
- viewpoint;
- scale; and
- spatial arrangement.
The task is consequently more than recognising isolated shapes. A useful vision system must often determine how objects are spatially and functionally related.
Depth Estimation
One of the central challenges of computer vision is estimating depth and distance from images.
A conventional camera image is normally two-dimensional, while the photographed environment is three-dimensional. The machine must therefore infer or measure the third dimension from available information.
Possible sources include:
- stereoscopic disparity;
- parallax between different viewpoints;
- motion;
- diminution of size;
- foreshortening and changing aspect;
- occlusion;
- texture gradients;
- known object dimensions;
- camera geometry; and
- active depth measurements from specialised sensors.
Depth analysis creates an important connection between human visual perception and machine vision, although the mechanisms employed by the two systems need not be identical.
Depth Cameras and Spatial Vision
A depth camera or depth sensor estimates the range of visible points within a scene.
Depth may be determined through passive stereo vision, active stereo vision, structured infrared light, time-of-flight measurement or related methods.
The resulting depth map associates image positions with estimated spatial distances. It can then contribute to object recognition, navigation, measurement or the construction of a three-dimensional point cloud or model.
Computer Vision and 3D Reconstruction
3D reconstruction is one of the major perspective-related applications of computer vision.
Several photographs of the same scene can be captured from different station points. Corresponding features then occupy different image positions because of parallax and changing perspective.
Photogrammetric and computer-vision methods can analyse these differences, estimate camera relationships and reconstruct three-dimensional geometry.
The result may be a depth map, point cloud, mesh, surface model or other three-dimensional representation from which entirely new perspective views can subsequently be generated.
Perspective Matching and Camera Calibration
A computer can also analyse an existing photograph to estimate aspects of the perspective system that produced it.
This may include recovering or estimating:
- the horizon;
- vanishing points;
- field of view;
- camera orientation;
- camera position;
- principal image directions; and
- other camera or projection relationships.
A computer-generated model or additional image element can then be aligned with the recovered perspective.
Perspective matching and camera calibration therefore demonstrate how principles developed within geometrical perspective remain highly relevant to modern digital imaging and computer vision.
Multi-View Computer Vision
A single perspective image contains only one view of spatial reality. Many computer-vision systems therefore make use of multiple views.
Images taken from different station points can reveal surfaces hidden in other views, provide parallax information and make it possible to cross-match corresponding scene features.
Multi-view image registration spatially transforms and aligns images within a common frame of reference so that information obtained from different views or sensors can be compared and combined.
This principle is fundamental to photogrammetry, 3D reconstruction and many advanced forms of computational imaging.
Motion and Optical Flow
Computer vision can analyse not only static images but the changing perspective patterns produced by movement.
Optical flow describes patterns of image motion associated with movement between an observer or sensor and the environment.
As a camera moves forwards, sideways or rotates, the projected positions and apparent sizes of scene features change systematically. Such changing relationships can provide information about direction, depth, relative movement and spatial organisation.
This makes motion perspective particularly important to robotic vision, vehicle navigation and other systems operating within changing environments.
Occlusion and Hidden Surfaces
Occlusion creates a fundamental problem for both human and computer vision.
A nearby object can conceal another object, while one part of an object can hide another part of itself. A single perspective image therefore contains only those surfaces visible from its particular station point.
Changing viewpoint can reveal previously hidden surfaces. Multi-view systems can consequently overcome some of the limitations of a single perspective image by combining information obtained from several directions.
Occlusion simultaneously provides a useful depth cue and creates missing visual information that an artificial vision system must interpret or overcome.
Facial Recognition
Facial recognition is a specialised form of visual identification in which a system analyses facial information in images or video.
The perspective problem is significant because the appearance of a face changes with viewing angle, expression, lighting, distance, occlusion and image quality.
A successful system must therefore compare useful visual features across a range of different perspective appearances rather than expect a face always to appear in one fixed form.
Robotic Vision
Robotic Vision refers to a robot’s or artificial system’s visual sensing and imaging system and is a particularly important application of computer vision.
A robot may need to determine:
- where it is;
- which objects surround it;
- where those objects are located;
- their size, shape and orientation;
- whether they are stationary or moving;
- which surfaces can be traversed;
- where obstacles occur; and
- how its own movement will alter the available perspective view.
Visual information can then be linked to mechanical action. A robot may identify an object, determine its spatial location and orientation, move towards it and manipulate it.
Robotic vision therefore makes the connection between perspective information and physical action especially explicit.
Autonomous Vehicles and Spatial Awareness
Autonomous and partially autonomous vehicles require continuous information about the surrounding spatial environment.
Camera and sensor systems can be used to detect roads, vehicles, signs, pedestrians and other objects while monitoring their relative positions and movement.
The system must repeatedly update its model of the environment as the vehicle itself moves. Every change of position produces a corresponding change of perspective.
Vehicle vision therefore combines motion perspective, object recognition, depth estimation, spatial measurement and real-time decision-making.
Computer Vision and Visual Navigation
Perspective information can also be used for orientation and navigation.
A moving camera observes continuously changing relationships between scene features. By comparing views through time or matching them against maps and models, an artificial system can estimate its location, direction of travel and relationship to surrounding objects.
This extends one of the basic functions of perspective—using visual information to understand position and direction—into automated machine navigation.
Generating Novel Perspective Viewpoints
Computer vision increasingly operates not only by analysing existing images but by helping construct models from which new perspective views can be generated.
Multi-view photographs, photogrammetry and computer-vision reconstruction can produce three-dimensional scene models. Once spatial geometry has been reconstructed, a virtual camera can be placed at a new position and a new perspective image generated.
This creates a powerful reversal and continuation of the ordinary image process:
Physical Scene → Multiple Perspective Images → Computer-Vision Reconstruction → 3D Model → New Perspective Images.
Vision-Language and Vision-Action Systems
Computer vision is also increasingly combined with language and action systems.
A Vision-Language Model can combine visual analysis with natural-language processing, allowing images and videos to be described, queried or related to textual information.
Related vision-language-action systems can connect visual interpretation and language instructions to physical actions. A robotic system might identify a bottle within a scene, interpret an instruction to pick it up, determine its position and then plan an appropriate movement.
Here perspective becomes part of a much larger chain:
View → Identify → Understand → Locate → Decide → Act.
Computer Vision versus Human Vision
Computer vision is sometimes described as an attempt to make machines “see”, but human and artificial vision should not be treated as identical systems.
Human visual experience involves the eyes, retinal projection, binocular vision, eye and head movement, visual attention and complex physiological and psychological interpretation.
A computer-vision system may instead use one or many cameras, depth sensors, LiDAR, infrared systems, databases, mathematical algorithms and machine-learning models.
Both systems can attempt to derive information about shape, size, position, depth, movement and identity, but the mechanisms through which this is achieved can be very different.
The application of perspective principles and theories to artificial intelligence’s understanding of vision therefore remains an important open field of investigation.
Computer Vision Methods and Systems
Computer vision encompasses a very large range of methods and systems. Common perspective-related tasks include:
- image classification;
- object and process detection;
- object identification and recognition;
- face recognition;
- edge and contour analysis;
- scene analysis;
- depth estimation;
- stereo vision;
- motion and optical-flow analysis;
- camera calibration;
- perspective matching;
- multi-view image registration;
- 3D reconstruction;
- robotic and autonomous-machine vision;
- image validation and certification;
- database image searching and comparison; and
- generation of novel perspective viewpoints.
Many modern systems use machine learning and artificial intelligence to identify useful patterns within very large quantities of visual data.
Applications of Computer Vision
Computer vision has applications across a wide range of scientific, technical and commercial fields.
- Robotics — perception, navigation and manipulation.
- Vehicles — spatial awareness, object detection and navigation.
- Manufacturing — inspection, recognition and automated processes.
- Security — image monitoring, identification and event detection.
- Traffic systems — analysis of vehicles, roads and movement.
- Medical imaging — analysis of CT scans, X-rays, endoscopic images and other medical data.
- Surveying and mapping — photogrammetry, image matching and 3D reconstruction.
- Drones — aerial imaging, mapping, inspection and environmental observation.
- Smart cities and the Internet of Things — machine interpretation of networked visual information.
- Image databases — classification, search and comparison.
- Virtual and augmented reality — spatial tracking, mapping and integration of real and digital imagery.
- Computer graphics and visual effects — camera matching, reconstruction and integration of photographed and generated imagery.
Computer Vision and Medical Imaging
Computer vision can analyse medical images produced by systems such as CT, X-ray and endoscopic imaging.
These images provide views of spatial structures that may be difficult or impossible to inspect directly using ordinary eyesight. Computational analysis can assist with identifying, measuring, comparing and classifying features within the resulting image data.
This illustrates the wider principle that computer vision is not restricted to ordinary photographic views. It can analyse images formed by many different optical, instrument and scientific imaging systems.
Computer Vision and Artificial Training Worlds
An artificial vision system can also be trained using computer-generated images and simulated three-dimensional environments.
Large numbers of artificial scenes can provide different objects, viewpoints, lighting conditions, spatial arrangements and unusual situations. The resulting perspective images can then be used to train artificial systems to recognise and respond to situations that may be uncommon in ordinary captured data.
Computer vision can therefore operate upon both images of physical reality and images generated from artificial spatial realities.
Limitations of Computer Vision
The extraordinary capabilities of modern computer vision should not obscure the fundamental difficulty of the task.
The same physical object can produce dramatically different images when viewpoint, distance, field of view, lighting, occlusion, motion, optical system or image quality changes. Different objects can also produce deceptively similar projected forms from particular viewpoints.
A single image may contain insufficient information to determine the original three-dimensional scene uniquely, while hidden surfaces may be unavailable altogether.
Computer vision therefore confronts many of the central problems of perspective: how spatial reality is transformed into an image, which properties survive that transformation, and how reliably those image properties can be used to reconstruct knowledge about the original world.
Computer Vision and Perspective Category Theory
Within Perspective Category Theory, a complete computer-vision system may involve several perspective categories rather than only one.
- Natural Perspective may provide the original physical spatial reality.
- Optical Perspective governs the formation of images through light.
- Instrument Perspective operates through cameras, depth sensors, scanners and other devices.
- Mathematical Perspective provides geometrical, coordinate and measurement procedures.
- New Media Perspective processes, analyses, connects, models and interprets digital visual information.
- Visual Perspective Type 2 becomes relevant wherever reconstructed or analysed images are subsequently viewed and interpreted by a human observer.
A typical computer-vision chain can therefore be represented as:
Physical Scene → Optical / Instrument Image → Digital Image → Computer-Vision Analysis → Measurement / Classification / Model → Decision, Representation or Action.
The system may accordingly combine several categories and functions of perspective within one continuous process.
Computer Vision — Frequently Asked Questions
What is computer vision?
Computer vision is a field of artificial intelligence that enables computers to analyse visual data and identify or interpret objects, scenes, events and processes represented in images and videos.
Why is perspective important to computer vision?
Most camera images are perspective projections of three-dimensional spatial reality. Understanding location, shape, scale, orientation, depth and movement therefore requires analysing how spatial properties have been transformed by viewpoint and projection.
What is robotic vision?
Robotic vision is computer vision used as part of a robot or artificial system’s visual sensing, interpretation and decision-making process, particularly where live images must guide real-time action.
What is the difference between computer vision and image processing?
Image processing can transform or analyse image data, while computer vision is particularly concerned with extracting useful information or interpretations about objects, scenes and processes represented by that data. Image processing can therefore form part of a computer-vision system.
Can computer vision determine depth?
Yes. Depth can be estimated or measured using information such as stereo disparity, parallax, motion, perspective depth cues, depth cameras, structured light and time-of-flight systems.
Can computer vision reconstruct a 3D scene?
Yes. Photogrammetric and computer-vision methods can combine several images, estimate camera relationships and reconstruct three-dimensional geometry as depth maps, point clouds, meshes or other spatial models.
What is camera calibration in computer vision?
Camera calibration and perspective matching determine or estimate properties of the imaging system and its relationship to the scene, allowing image measurements and multiple views to be related more accurately to spatial geometry.
What is optical flow?
Optical flow is the pattern of image motion produced by relative movement between an observer or camera and the surrounding environment. It can provide useful information about movement, direction and spatial structure.
Is computer vision the same as human vision?
No. Both can obtain information about spatial reality from images, but human vision involves biological, physiological and psychological processes, while artificial vision may use cameras, sensors, algorithms, mathematical models and machine learning.
What can computer vision recognise?
Depending upon the system, computer vision can classify and identify objects, faces, scenes, visual features and processes, estimate spatial relationships and movement, compare images and support navigation or physical actions.
What is a Vision-Language Model?
A Vision-Language Model combines computer vision and natural-language processing so that visual information can be analysed in relation to language, enabling tasks such as image description, visual question answering and broader image analysis.
Computer Vision within the Wider Field of Perspective
Computer vision demonstrates why perspective has become increasingly important far beyond the traditional fields of drawing and representation. A machine attempting to understand an image confronts many of the same fundamental relationships studied throughout perspective: viewpoint, direction, projection, size, shape, depth, distance, orientation, occlusion, movement and changing appearance.
The field also extends perspective from the production of images towards their systematic measurement, matching, classification, reconstruction and interpretation. Perspective becomes not simply a way of depicting spatial reality, but a means of extracting knowledge from its views.
Computer vision is therefore one of the most important contemporary meeting points between perspective, optics, imaging, mathematics, artificial intelligence, robotics and three-dimensional spatial modelling.
You must be logged in to post a comment.