Automatic focusing method and system based on convolutional neural network

Abstract

The invention relates to an automatic focusing method and system based on a convolutional neural network, wherein the method includes the following steps: establishing a defocus image data set; building a defocus estimation model, and using the defocus image data set to estimate the defocus amount The model is trained; the current frame image collected by the image detector is obtained, and the current frame image is input to the trained defocus estimation model to predict the image defocus amount, and the defocus amount relative to the current focus motor position is normalized Calculate the defocus value according to the normalized output value of the defocus amount, and judge the adjustment direction of the focus motor; adjust the position of the focus motor according to the defocus value and adjustment direction, and complete the image detector. Auto focus. The invention can automatically extract the defocus amount of the input image by using the trained defocus amount estimation model, effectively avoiding the problem that the artificial defocus amount feature is difficult to extract, the focusing process is simple and fast, and can be used for continuous changes in object distance Continuous focusing occasions.

Description

technical field [0001] The invention relates to the field of imaging technology, in particular to an automatic focusing method and system based on a convolutional neural network. Background technique [0002] Clear imaging is an important prerequisite for the normal operation of imaging equipment, and inability to focus accurately is the primary reason for affecting imaging clarity. In order to quickly and clearly image the targets in different positions and different moving states, the optical system must realize real-time and fast automatic focusing. [0003] Autofocus methods for imaging devices can be divided into two categories: ranging methods and image methods. [0004] The ranging method mainly uses ranging devices such as laser rangefinders to measure the distance of the observed target (ie, the object distance), and adjusts the optical parameters according to the conjugate relationship of the object and image to achieve fast focusing of the system. The disadvanta...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that additional auxiliary mechanisms such as distance measuring devices need to be added, which increases the complexity and structural cost of the system.

This type of method is more suitable for continuous automatic focusing occasions where the object distance changes continuously, but there is a problem of difficulty in extracting artificial defocus features, which limits the engineering application of this method

Chinese patent CN106249508B proposes an automatic focusing method, which first extracts face blocks in the field of view, and inversely deduces the object distance information according to the proportional relationship between the face block area and the actual face area, and then according to the conjugate relationship of the object image , to adjust the optical parameters to achieve automatic focusing. The method described in this patent is fast and effective, but it is only suitable for occasions where there are human faces in the field of view, and its applicability is narrow

Images