|The ISRG conducts interdisciplinary research into the theory and application of image processing and image understanding techniques. The center's projects include automatic pavement analysis, knowledge-based image segmentation, stereo vision, visualization and fractal-based image processing.|
|Multiple Object Recognition|
|Participants: Mark R. Stevens
Previous work on model-based object recognition has focused on the recognizing single objects in an image. Unfortunately, objects rarely exist in isolation but rather are part of a complex scene. Moving beyond recognizing one object at a time allows for reasoning about each object's relative placement within the scene. Problems such as occlusion, and diffuse-diffuse color interactions can now be handled explicitly. In order to locate multiple objects simultaneously, an iterative approach based on using computer graphics techniques refines a given scene hypothesis in order to find the best interpretation of the objects lying before the camera. The method consists of three stages: 1) render a prediction of the scene based upon the hypothesized configuration of 3D objects, 2) pixel-wise match the rendered image to the sensor data, 3) refine the scene parameters to bring the rendered and observed images into alignment.
Participants: Mark R. Stevens
Recognizing military vehicles in outdoor scenes is a daunting task for a computer algorithm. We are looking at iterative methods for refining the pose of an object based on information fused from color, thermal and range sensors. The project is a part of the UGV (Unmanned Ground Vehicle) project through DARPA. We have had good results recognizing M60, M113, M113-901 and pickup truck models in realistic ATR imagery. In many cases our algorithms are able to cope with large amounts of terrain occlusion and structured scene clutter. The occlusion is dealt with by inferring occluding surfaces in the range data and using that information to remove features being sought from the model.
firstname.lastname@example.org / Thu Oct 14 18:51:28 EDT 1999