Matching method of parallel multi-ocular image for corn canopy

A image matching model of parallel multi-ocular is introduced. The camera is rectangle arranged four channel multi-spectral camera, the four channels is R, G, B and NIR respectively. The corn canopy images in field are captured by the camera, and the distance between the camera and the corn canopy is about 0.5 m. The features of the images are analysed, and a new matching method of images is proposed. One channel image act as source, others as destination, and the edge points of the corn leaf in source image are source feature points, feature vector of each point is composed by 18 directional derivatives. After that, the destination feature point is searched in destination image. In a local area of destination image, if the intersection angle of edges that point and source point located is smaller than one threshold, and the distance between feature vector of the point and source feature vector is minimum among all of them, then the point is matched destination feature point, and the feature points pair set is constructed. Each five non-co-linear feature point’s pair in above set can construct one affine transform. Averaging all such affine transforms, the affine transform from destination image to source image is constructed, and the match of two images is finished. This method is suitable for irregular object, complicated background, and the object change a lot in different channel.