Skip to main content

Real-Time Generation and Display of 2.5-D Sketches for Moving Scenes

Objective

The GENEDIS project aimed to develop and implement a demonstrator prototype of an imaging system which is capable of producing a form of 2.5-D sketch directly. This sketch, in which image intensity was to be related to range, was intended to be in one-to-one pixel correspondence with an illumination intensity representation generated simultaneously. Statistical operations performed on successive range estimates were expected to show a marked improvement over the single measure resolution.
The project aimed to develop and implement a demonstrator prototype of an imaging system capable of producing a form of 2.5 dimensional sketch directly. At the conclusion of the project conception phase, it was decided that the system should be composed of 4 parts: 2 cameras arranged to deliver 2 images of the same scene; 2 feature extractors for extraction of relevant features in both images (an edge-based stereopsis approach based on the Moravec operator was chosen); a correlator, based on dynamic programming, which detected the corresponding features in both images and computed the range of the imaged point in the viewfield from their disparity; and an interpolator, which interpolated range from the ranges delivered by the correlator, and allowed determination of the range of all the points in the viewfield. A successful demonstration of the integrated system (except the interpolator) was presented. The 3 main components, the aligned twin cameras, the real time edge extractors, and the pipeline real time correlator could be reused in a variety of systems for pattern recognition purposes. A breakthrough in the speed of measurement of scene depth matched the pace of interpretation to changes in the scene from stereo images. Real time applications, such as the identification of grasp points for robots, are now possible.
At the conclusion of the project conception phase, it was decided that the system should be composed of four parts:
-two cameras arranged to deliver two images of the same scene
-two feature extractors for extraction of relevant features in both images (an edge-based stereopsis approach based on the Moravec operator was chosen)
-a correlator, based on dynamic programming, which detected the corresponding features in both images and computed the range of the imaged point in the viewfield from their disparity
-an interpolator, which interpolated range from the ranges delivered by the correlator, and allowed determination of the range of all the points in the viewfield.
As the result of the first phase of the project, a successful demonstration of the integrated system (except the interpolator) was presented. The three main components, the aligned twin cameras, the real-time edge extractors, and the pipeline real-time correlator are attractive features that might be reused in a variety of systems for pattern recognition purposes.
A breakthrough in the speed of measurement of scene depth matched the pace of interpretation to changes in the scene from stereo images. Real-time applications, such as the identification of grasp points for robots, are now possible.
Exploitation
A prototype of a low-priced system to deliver range images for industrial scenes is under development. The expected performances are 2 mm error at 1 metre for a 512 x 512 pixels image for a system working at video rate and 1.3 mm of lateral resolution. Atthe moment no similar low-priced system exists on the market. Further developments to upgrade the system for image processing will be necessary prior to industrialisation.

Coordinator

University of Strathclyde
Address
Richmond Street 26
G1 1XH Glasgow
United Kingdom

Participants (2)

Barr and Stroud Ltd
United Kingdom
Address
Anniesland Caxton Street
G13 1HZ Glasgow
Zeltron SpA
Italy
Address
Via Principe Di Udine 66
33030 Campoformido