Computer Vision

Computer vision really is just what it says - giving a computer the ability to visually process its surroundings. Computer vision is taking a computer or a system of computers and a camera or network of cameras to acquire images, transform and interpret those images, and extract various types of useful information from images and/or image streams as in the case of video. There are various applications of computer vision in engineering research. Automatic segmentation of a bone from a CT scan and the analysis of x-ray video to determine hip implant separation are two examples of computer vision applications to the medical field. Automatic obstacle avoidance for an autonomous vehicle or the localization and mapping of the immediate surroundings to a vehicle are examples of computer vision in the robotics field.

Image Processing and Image Understanding

The difference between image processing and computer vision is a very subtle distinction, and in fact many researchers may define them differently than I have here. Where computer vision is the application of the computer understanding its environment, image processing would be the methods used to take an image and transform, segment, determine properties, or perform other processing and analysis tasks to aid in the computer vision. Image understanding is related to computer vision and is an integral subset of computer vision which specifically involves aiding the computer application to gain a deeper understanding of the camera's surroundings and activities within the field of view. An example would be to interpret the gestures of an individual to control a large construction machine, or to classify the items of an image as grass, trees, rocks, or road for a vehicle navigation problem.


Using a mobile platform of any kind can be considered a type of robotics application. There are many ways to automate vehicles in order to provide local environmental awareness behaviors such as bump or range sensors. Vision is still a rarely used conduit for environmental information to flow through. There are several reasons for this such as processing power and applied robust understanding algorithms. It is the goal of many current computer vision researchers (myself included) to provide ubiquitous solutions to these problems. Some of my recent applications use techniques which can track a camera pose solely and directly from the image data as the camera moves through a scene. Ideally this would be attached to a robot and provide navigation information to a path planner which would give the robot a visual understanding of its surroundings.

Medical Imaging

I have recently been involved in medical image processing to compute 3D volumetric data sets from video x-ray data. Specifically, my master's thesis work has included research in the processing of fluoroscopy data generated along an arbitrary path to reconstruct three dimensional models much like that obtained by CT or MRI. I have also been involved in other projects for medical imaging including general segmentation of medical images, automatic detection of hip implant separation in fluoroscopy video, and metal artifact reduction techniques in CT scans. Much analysis in the medical field relies heavily on visual understanding to diagnose a patient. Any automation which could aid a physician in a faster more accurate diagnosis would be beneficial to society in general.