Automatic tracking method and device based on visual feature points

The present invention provides an automatic tracking method based on visual feature points. The method comprises collection of image information and depth information, wherein the collection moments of the image information and the depth information are in one-to-one correspondence. Employing a sharpening processing and weight equalization processing, and determining prospect feature points; estimating the estimation positions where prospect features may occur at a second frame moment according to the feature points of a first frame image; matching and fusing the feature points of the second frame image and the prospect feature points of the estimation positions to obtain the prospect feature points with determined positions; estimating the amplifying and/or rotation coefficients of the feature points of the second frame image according to the prospect feature points with determined positions, and performing threshold determination to obtain valid feature points; and detecting the depth distance distribution,