Abstract
The complexity of many computer-recognition problems is such that speed of processing is an important factor that cannot be overlooked, especially when real-time applications are sought. In order to reduce the processing time of matching, our work is implemented in parallel on SIMD architecture. We describe an extension of the Holder and Buxton algorithm (1989) using the surface normal and axis of rotation of objects as ‘edge’ features to increase the object-recognition scope to objects containing developable surfaces. In addition, we implement an improved data-sorting algorithm that gives impressive speed ups compared with the earlier sorting technique. We show the method to be highly effective in the fast determination of scene interpretations with tests using artificial scenes generated efficiently by a parallel ray caster incorporating constructive solid geometry (CSG). Accuracy and robustness are further tested by application to a real-world scene.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call