Fuzzy logic helps match patterns fast
Classification and recognition of objects in video images is a difficult task involving complex algorithms and training techniques. To address this, start-up BrainTech (North Vancouver, BC, Canada) has developed a PCI-based system that performs both recognition and tracking of objects in video frame times. According to Paco Nathan, BrainTech`s chief technology officer, the first customers for the system are expected to be in high-speed web-inspection applications.
Consisting of a color video camera, frame grabber, a BrainTron processor board, and software for image preprocessing and object classification, the system first performs background removal, noise filtering, and boundary detection. Image data are then extracted from the video signal for training and recognition by the processor, a board that uses a fuzzy-logic-based adaptive vector quantizer to classify and recognize objects. Fuzzy results from the processor are then integrated with position tracking data through a statistical classifier that identifies moving objects in the video frame.
To perform pattern matching at 30 frames/s, the vector quantizer has a parallel, pipelined architecture. When a single processor is used for pattern recognition, the pattern-matching database is stored on a single device. As input data arrive, the processor performs all the calculations needed to match input data against stored patterns.
The system has been demonstrated for handwriting recognition. In operation, the system is trained to associate handwriting samples with computer characters. After training, the system automatically recognizes characters closest to the samples. BrainTron can be reached at http://www. bnti.com.