Optical temperature measurement technique has been widely researched in current temperature detection field. Because the three-color thermometry based on color CCD has many disadvantages affecting temperature measurement precision, such as the non-ideal spectral responsive bandwidth of RGB three-channel, the emissivity variance and the nonlinear output channel, as well as the few visible light share of all the radiant energy. So a new design of temperature measurement system based on two-color imaging in adaptive optics of CCD has been proposed in this paper. This system was designed according to the principle of self-adaption imaging based on the feedback error signal. Especially, the optical passive athermalisation was applied to eliminate the impact of temperature change on system image quality, and the substrate-mask was also placed in optical path to resolve the problem of external stray light interference. Furthermore, a new temperature measurement algorithm of two-color radiation imaging based on grayscale inversion was deduced, which can avoid the influence of gray body assumption and non-ideal spectral response bandwidth, and correct the interference of CCD hardware noise as well. Another fast registration algorithm of two-color image based on wavelet transformation and mutual variance was also introduced, which has a high translation and rotation registration precision. Eventually, the field experiment has been carried out based on this temperature measurement system. And the result has shown that this system can calculate and display the temperature distribution of boiler flame, and have the characteristics of high accuracy, good real-time and strong practicality.