Computer Vision Research
Site Menu
Introduction | White Paper

A White Paper
3D Computer Vision


Several techniques have been developed to extract accurate three-dimensional information from the range of calibrated video input devices. The current research is focused on making the computer vision system simple, accurate, and fast. All algorithms are developed for multiprocessor architectures. Most of the algorithms use only very simple mathematical calculations. These and some other techniques make possible to develop accurate real-time 3D computer vision system using only equipment available on the common market.
The technology is unique in that it utilizes an intelligent digital video processing technique in which both measurements of data and 3D visualization of objects are obtained on a real-time basis without human intervention.


The three dimensional information is being calculated from a range of calibrated video input devices. The algorithm is based on exactly known positions and optical parameters of video input devices. Three-dimensional acquisition process can be divided into two major steps: feature detection and matching. First, the feature detection process is performed on each frame acquired from all input devices. Then the matching algorithm is performed to track each feature element from frame to frame on all video input devices. The three-dimensional information is being calculated based on tracked paths of features.


Automatic Real-Time Spatial Information Acquiring

All processes are performed automatically. No operator intervention is necessary in all steps of acquiring and processing the spatial information. All processes are performed in real-time manner.

Generating Models Based on Acquired Spatial Information

The acquired data can be analyzed for building the mathematical approximation of all visible objects surfaces. This surface description is used for generating textural information and modeling.

Object Recognition

Several techniques have been designed to recognize the objects in the captured three-dimensional data. The object recognition process is performed in a real-time manner. Accurate 3D data allows achieving very low recognition error.

Object Tracking

After object has been recognized it can be easily tracked. The exact spatial position, speed, and direction of object motion can be determined in a real-time manner.

Using Electromagnetic Waves Other than Visible Light

Besides of the electromagnetic waves in the visible light spectrum, other waves can be used. Using infrared spectrum allows developing night three-dimensional computer vision system. Using high frequency electromagnetic waves (X-Rays) allows acquiring thee-dimensional information inside other objects, such as human skeleton.



The computer vision system has large scale of applications in robotics. Visual feedback is essential part of almost any robotic system. "Blind" robotic systems suffer from many problems such as increasing operational error, sensitivity to accuracy of given scene, etc. To eliminate these errors very expensive and complicated methods are being used. The most easy and evident approach is developing robots that can simply "see and understand what they do".


The computer vision system can be used as an additional tool for Computer Aided Design (CAD) applications. The system can be designed as a hand-held device. The operator can just walk inside or around objects with such device. The system will acquire, match, and record all spatial information around. In a very short period of time operator will have full real three-dimensional information about the desired object. The three-dimensional information can be easily ported into popular CAD applications for further processing.

Security and Surveillance

The computer vision system is very useful in the security and surveillance systems. The current "two-dimensional" video surveillance systems use human personnel to watch and analyze the scene. These systems suffer from high error rate caused by human factor. The computer vision systems watch and analyze the scene and automatically generate "pre-alarm" situations. A human personnel is used only for making the final decisions.


The computer vision system can be used as a very easy-to-use and accurate measurement device. The operator just has to "put in front of the camera" any object. In fractions of seconds the system will determine all spatial information of given object, such as size, volume, minimal bounding box, etc.


The proposed system can be used to detect and track objects. The moving vehicle is an example. The system can detect and record the speeds and directions of all vehicles on a given part of road simultaneously. Since the computer vision systems are passive, no device can detect the presence of such system. So developing any "anti-radar" devices will be impossible.

Copyright (C) 2006-2009 Andre Mirzoyan