摘要 |
PROBLEM TO BE SOLVED: To recognize an object in a moving image with high accuracy. SOLUTION: A characteristic point array in a database 5 is corrected with a hypothetical set 7, an input frame is collated with a corrected database template by the variable template matching of a multiple resolution Gabor wavelet, positional deviation is estimated by the phase of a wavelet coefficient, and an object in a video frame is tracked. Matching results surpassing a threshold are used to correct a hypothesis in the set 7 or to generate a new hypothesis. Only hypotheses whose evidence values are high in the set 7 are smoothed by temporal and spatial filtering 4 about all hypotheses after matching of them with an input frame is finished, and the smoothed hypotheses are outputted as recognition results.
|