Accession Number : ADA322958

Title :   Outdoor Landmark Recognition Using Hybrid Fractal Vision System and Neural Networks.

Descriptive Note : Final rept. 1 Oct 93-Dec 96,

Corporate Author : NORTH CAROLINA STATE UNIV AT RALEIGH DEPT OF ELECTRICAL AND COMPUTER ENGINEERI NG

Personal Author(s) : Luo, Ren C.

PDF Url : ADA322958

Report Date : 24 FEB 1997

Pagination or Media Count : 51

Abstract : The objective of this research project is to develop a hybrid vision system to isolate and recognize landmarks in outdoor natural scenes. We have used two-steps approach to solve this problem. The first step is to isolate the landmarks based on image segmentation using fractal models for objects in the images, and the second step is to recognize the isolated landmarks using neural networks. An Incremental Fractional Brownian Motion(IFBM) model for segmenting and isolating the outdoor natural images have been developed and successfully tested. Two neural networks systems namely, a reconfigurable multilayered feed forward neural network and self organizing neural network, with adaptive learning capabilities have been designed and trained on a variety of different outdoor traffic signs. The networks are robust and efficient and can recognize signs successfully with vigorous test. We have implemented this hybrid system onto an autonomous mobile robot to achieve vision based navigation with natural landmark recognition. The system has been successfully tested in the lab environment laid-out with a variety of traffic signs. The robot can navigate through scattered environment with random obstacles and traffic signs. The robot has won twice championship on international autonomous mobile robot research and application competition in 1993 and 1995, respectively. The contenders consists of universities, government laboratories and companies from both US and outside of US. The competition was sponsored by AAAI ( American Association of Artificial Intelligence). The results of this research can be extended to a variety of both military and commercial applications such as military surveillance, autonomous vehicle navigation, automatic target recognition, indoor autonomous service robot navigation and object inspection.

Descriptors :   *IMAGE PROCESSING, *NEURAL NETS, *COMPUTER VISION, DATA BASES, FRACTALS, ROBOTICS, TARGET RECOGNITION, ROBOTS, LEARNING MACHINES, COMPUTER APPLICATIONS, PATTERN RECOGNITION, ARTIFICIAL INTELLIGENCE, SYSTEMS ANALYSIS, AUTONOMOUS NAVIGATION, BROWNIAN MOTION, HYBRID COMPUTERS, SELF ORGANIZING SYSTEMS.

Subject Categories : Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE