High-performance Digital Camera Auto-exposure Algorithm Research And Implementation

This thesis explores the theory of auto-exposure (AE) algorithms of digital cameras. It develops two novel algorithms based on evaluation results of existing ones.The basic theory of digital imaging system is first explained. It includes the two successive optical and electrical signal pipelines during the image capturing process, the standard of AE algorithms, three factors which affects exposure and the typical execution flow of the AE algorithms.After searching and analyzing the existing AE algorithms, this thesis concludes the problems with these algorithms and presents two novel ones accordingly: one is based on the direct-setting method which shortens the exposure execution time under normal lighting conditions. With the brightness of the current image and the exposure parameters, it obtains through a derived equation the exposure parameters that result in the reference brightness within the next frame of image. The other algorithm uses dynamic-partition method to compensate the main object under special lighting conditions. It takes advantage of the large brightness contrast between different elements in the image and dynamically distinguishes the main object and the background. The compensation is made to the reference brightness according to the contrast between these two regions. This algorithm is capable of handling both special lighting conditions at the same time and minimizing the influence of changing object position. The above two algorithms are both developed and simulated on the AE evaluation platform built during the process of this research.This thesis then presents about the systematic implementation of these two algorithms. A software and hardware cooperation mechanism is adopted, with the hardware responsible for the real-time image statistics calculation and the software controlling the algorithm execution flow. The hardware is implemented on FPGA and the software on ARM. An AE&AWB statistics engine together with a series of controlling strategy is successfully developed in the purpose of minimizing operation circuit and controlling logic units.A test platform is constructed to verify the implemented algorithms. A great amount of valuable data is obtained base on the test results of various combinations of scenes and light sources. The algorithms are proved to meet the expected performance. At last, this thesis predicts some possible improvements on the proposed AE algorithms.