Study On The Application Of Autofocus Technique Based On Imageprocessing Technology In Aerial Camera

Real-time accurate autofocus technique is a key technology to enhance the ability of aerial cameras to produce high-resolution images. Due to the complexity of aerial environment, the function of aerial cameras with a long focal length is affected by many factors(such as temperature, pressure, and change of imaging distance) which may result in defocus and degrade image resolution. Therefore, it is necessary to develop an autofocus system for aerial cameras. Because of the low speed and poor accuracy of the automatic calibration methods, more and more attention has been attracted by the autofocus technique based on image processing technology. Following this trend, this work aims to solve several problems that one may encounter when using this technique in aerial cameras, including selection of focus measure function, single noise ratio, image motion, and improvement of autofocus precision.Based on the detailed analysis of previously proposed focus measure functions, a general focus measure model is established. By comparing the performances of those focus measure functions, a focus measure based on wavelet transform is selected to be used in aerial cameras. To improve performance, the selected function is modified. The modified function is featured by strong ability of noise reduction and highly sensitive measurement of focus length, which makes it suitable for application in aerial cameras.A method to analyze quantitatively the relationship between SNR and autofocus precision is proposed. In addition, the mathematical expression of their relationship is given. The variance of focus measure functions and the autofocus uncertainty measure function(AUM) are used to measure the autofocus precision. The relationship curve between SNR and autofocus precision is obtained by calculating the variance of focus measure functions and AUM. When the SNR of an image changes from 29 dB to 62 dB, the AUM decrease from 0.1428 mm to 0.0391 mm by using the modified focus measure function based on wavelet transform. The accuracy of autofocus increased by about 30%.The influence of image motion on autofocus precision has been analyzed. The larger the image motion is, the more the peak of focus measure declines and the lower the autofocus precision is. A method is proposed to lower image motion, i.e., moving the detector quickly and imaging continuously. The influence of motions can be weakened by using the method of quadratic fitting. The peak position of the quadratic fitting curve gives the optimum focus position. If the motion is less than 3 pixels the autofocus precision can be improve by approximate 40% through the above method, meeting the standard of aerial cameras.A method in which an image is divided into many sub-regions for autofocus is proposed to improve the speed and precision of autofocus. In this method, parallel calculation is used and useless sub-regions are deleted. Consequently, the autofocus precision is enhanced, the speed autofocus is improved, and the real-time autofocus of aerial cameras is realized.The results obtained in this study enable the application of autofocus based on image processing technology to move forward.