The atmospheric coherence length which reflects the intensity of atmospheric turbulence is a very important parameter of laser atmospheric propagation and adaptive optics. It has been used as the modern definition of atmospheric seeing in astronomical observations. Day-night atmospheric coherence length monitor is a conventional instrument which can measure the atmospheric coherence length. It uses the two aperture differential imaging motion method (DIMM) to measure the atmospheric coherence length r0 based on continuous observation of stars. When there are clouds in the sky especially for cloudy weather, the starlight are often obscured by clouds and the SNR of image is very low for only faint starlight received by Day-night atmospheric coherence length monitor. This would cause unacceptable measurement error. Therefore, there are fewer atmospheric coherent length data in the cloudy weather. The paper analyzes the measurement errors and tracking errors under the low SNR condition. Based on the experimental data, the characteristics of night cloudy sky background are analyzed. According to the fact that the sky background obeys Gauss distribution, we present an improved method of Gaussian spot extraction fitting method to extract the optical spot region. The processing method includes the following steps: first, the mean intensity of the image is calculated by Gauss fitting, then, divide the image into the background and the suspected target with the threshold of the half width of the Gauss distribution above the mean intensity. The largest area of the suspected target will be taken as the optical spot area. After that, a sliding window of 5 multiply 5 scale is used to scan the optical spot area to carry out gauss fitting for the center pixel xij at the angle of 0degree,45degree,90degree and 135degree. A comparison is taken between xij and the average values of the four fitting results, if the difference exceeds the threshold value, the pixel xij will be taken as the noise pixel, and the mean value of the fitting result is used to replace the pixel intensity. In order to verify the accuracy of the method, the turbulence phase screen has been used in numerical simulation. The result shows that this barycenter extraction method can accurately extract the centroid of optical spot under low contrast situation. Even when the SNR is 1.25db, the measurement error of r0 is still less than 10%. We believe that this method may be useful in improving the adaptability of Day-night atmospheric coherence length monitor to more kinds of weather.