SBIR-STTR Award

This proposal describes a computational model of human observers, DYTA(dynamic Target analysis), for dynamic target detection, depth and velocity estimation DYTA will be developed based on human perception knowledge as well as physics of moving vehicles. The computational model consists of two levels of processes, early and cognitive visual processes. At the early visual processes level, DYTA has two independent units: Gabor multi-channel filtering to model human early vision processing, and computation of motion vectors of moving vehicels based on the physics of moving vehicles. At the cognitive level, DYTA has a fuzzy learning and inference model for reasoning about the behavior of dynamic targets, such as Location, distance and velocity. The algorithma we will develop in phase I includes computing target location by combining texure and motion features; computing depth and velocity of moving targets using rigid body motion techniques; and performing fuzzy learning and inference. A number of basic research issues will be addressed including moving vechicle detection, 3D structure recovery from image sequences,robust fuzzy learning and inflerence under uncertain and incomplete knowledge. DYTA is designed to predict human observer's response to the dynamic information in a scence,in particular, to moving vehicle location, distance and velocity. Potential Commercial Application: This technology has potential applications in both military and commercial markets. Examples of such applications are military dynamic target detection and acquisition, military and commercial surveillance, a monitoring and warning system in military operational driving, commercial vehicle saftey study, driver's warning system and visual enhancement, highway traffic control. etc

VisiTek proposes to develop an Intelligent Sensing System for Obstacle Detection and Avoidance(ISSODA) for robotic vehicles. ISSODA consists of a suite of fully integrated passive sensors and an obstacle detection and avoidance software package applicable for a variety of military and civilian tasks. The sensor suite consists of a triple CCD cameras for accurate 3D computation, a color video camera for obstacle detection, classification, and motion detection, a stereo FLIR sensors to allow operations in low-visibility environment. The software contains two major components, depth computation, obstacle detection and avoidance. The depth computation component consists of three algorithms, trinocular stereo vision, motion+stereo, and a fast stereo processor. The trinocular stereo vision and motion+stereo algorithms attempt to reduce stereo vision error and increase the accuracy of range data. The obstacle detection and avoidance component consists algorithms for color clustering, computing depth variance and motion variance, texture segmentation, moving obstacle detection, and a neural network classifier. ISSODA will be designed in such a way that its constituent parts (the various sensors, and algorithms) are all easily interchangeable. ISSODA will be a fully integrated system ready to be mounted on a mobile robot for sensing obstacles, sending warning signals, or generating a 3D grid map with obstacles marked and elevated.

Keywords:
OBSTACLE DETECTION, TRINOCULAR STEREO VISION, COLOR IMAGE SEGMENTATION, MULTI IMGERY INTEGRATION, OBSTACLE AVOIDANCE, DEPTH FROM STEREO AND M