Image analysis – Pattern recognition – Template matching
Reexamination Certificate
2000-12-01
2003-12-02
Tran, Phuoc (Department: 2621)
Image analysis
Pattern recognition
Template matching
C382S154000, C382S165000, C382S170000, C382S219000
Reexamination Certificate
active
06658150
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to an image recognition system for recognizing the correspondence between object images contained respectively in two images obtained by two imaging devices.
2. Description of the Relevant Art
Conventionally, a method of detecting a distance to an object is widely known, in which two imaging devices are positioned apart from each other, and a distance to the object is detected based on images of the object contained in two images produced respectively by the two imaging devices. In this method, proper recognition of the correspondence between the object images contained in the respective two images is a prerequisite to accurate detection of the distance to the object.
The SAD (Sum of Absolute Differences) method is a well known method in which one of two images is set to a reference image, and the other of them to a comparison image, and an object image in the comparison image which corresponds to an object image in the reference image is extracted. In this SAD method, the absolute value of a difference between the luminance value of each pixel in a reference pattern which is contained in the reference image and containing the object image, and the luminance value of a corresponding pixel in each pattern which is contained in the comparison image and identical in shape to the reference pattern is calculated on a pixel-by-pixel basis, and the sum of the absolute values is set to a correlation value ERR indicative of the correlation between each two patterns, and then, a pattern minimizing the correlation value ERR (i.e. a pattern having a highest correlation with the reference pattern) is extracted as a corresponding pattern which corresponds to the reference pattern.
However, when the SAD method is employed, if the two imaging devices are different in characteristics (e.g. if the lens of one of the imaging devices is dirty and the luminance values of the corresponding image are generally lowered), the difference between the luminance values of the two object images contained in the respective reference and corresponding patterns is increased even though the two object images are identical in shape to each other, which results in increasing the correlation value ERR and in worst cases, the two object images cannot be determined to be identical.
To solve this problem, a stereo imaging system was proposed in Japanese Laid-Open Patent Publication (Kokai) No. 10-255020, in which a sensitivity-adjusting device is interposed between the two imaging devices and an image recognition system that carries out detection of a distance to an object and other functions, whereby levels of signals output from the respective imaging devices are adjusted for compensation for the difference in characteristics between the two imaging devices.
The sensitivity-adjusting device of the stereo imaging system stores data conversion tables associated respectively with the two imaging devices in a memory, and levels of the signals output from the imaging devices are converted by using the data conversion tables to output the resulting data. Further, to cope with the aging of the sensitivity-adjusting device, values of the data conversion tables are updated by using image data of a specific object at a location within the scope of the common imaging area of two imaging devices.
However, in the above conventional stereo imaging system, it is required to provide the sensitivity-adjusting device between the imaging devices and the image recognition system as well as the memory for storing the data conversion tables. This complicates the construction of the whole system, resulting in a considerable increase in manufacturing costs.
The adverse effects of the difference in characteristics between the two imaging devices can be eliminated by using a normalized cross-correlation parameter as a correlation parameter in place of the correlation value calculated by the SAD method. In the method using the normalized cross-correlation parameter, however, a large amount of arithmetic operation is required, which puts an increased load on an operation system. Therefore, when the image recognition system is installed e.g. on an automotive vehicle for monitoring environmental conditions, it is required to use an expensive arithmetic unit having a high operation speed, which inevitably becomes a main factor of increased manufacturing costs of the automotive vehicle.
SUMMARY OF THE INVENTION
It is an object of the invention to provide an image recognition system which is capable of accurately determining identity between object images contained respectively in images obtained by two imaging devices, with hardware having a relatively simple construction and through a small amount of arithmetic operation, even when the two imaging devices are different in characteristics.
To attain the above object, the present invention provides an image recognition system for recognizing correspondence between object images contained respectively in two images obtained by two imaging means, the image recognition system comprising:
corresponding pattern-extracting means for setting one of the two images to a reference image and the other of the two images to a comparison image, calculating a correlation parameter indicative of a correlation between a reference pattern contained in the reference image and containing an object image, and each pattern contained in the comparison image and identical in shape to the reference pattern, by using a luminance value indicative of a luminance of the reference pattern and a luminance value indicative of a luminance of each of the patterns contained in the comparison image, and extracting, based on the calculated correlation parameter, a pattern having a highest correlation with the reference pattern from the comparison image, as a corresponding pattern which corresponds to the reference pattern; and
identity-determining means for determining identity between the object image contained in the reference pattern and an object image contained in the corresponding pattern based on a difference between a contrast value indicative of a contrast of the reference pattern and a contrast value indicative of a contrast of the corresponding pattern.
According to this image recognition system, the correlation parameter indicative of the correlation between the reference pattern contained in the reference image and containing the object image and each pattern contained in the comparison image and identical in shape to the reference pattern is calculated based on the luminance values of the two patterns, and then a pattern having a highest correlation with the reference pattern is extracted from the comparison image based on the calculated correlation parameter, as a corresponding pattern which corresponds to the reference pattern. Then, identity between the object image contained in the reference pattern and an object image contained in the corresponding pattern is determined based on a difference between a contrast value indicative of a contrast of the reference pattern and a contrast value indicative of a contrast of the corresponding pattern. Therefore, it is possible to accurately determine identity between the object images contained in the images obtained by the two imaging devices, respectively, with hardware having a relatively simple construction and through a small amount of arithmetic operation, even when the two imaging devices are different in imaging characteristics.
Preferably, the identity-determining means includes contrast difference component-calculating means for calculating a correlation parameter component corresponding to the difference between the contrast value indicative of the contrast of the reference pattern and the contrast value indicative of the contrast of the corresponding pattern, as a contrast difference component, and determination means for determining identity between the object image contained in the reference pattern and the object image contained in the corresponding pattern, based on a parameter obtai
Hattori Hiroshi
Nagaoka Nobuharu
Tsuji Takayuki
Watanabe Masahito
Blackman William D.
Carrier Joseph F.
Carrier Blackman & Associates P.C.
Honda Giken Kogyo Kabushiki Kaisha
Tran Phuoc
LandOfFree
Image recognition system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image recognition system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image recognition system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3102638