In order to generate true random sequences, a random number extraction algorithm with variable frame rate was proposed by using laser speckle video of atmospheric turbulence as entropy source of random number generator. Firstly, an image sampling method with variable frame rate was proposed to reduce the effect on randomness of high correlation between adjacent laser speckle images. Secondly, random numbers were extracted by processing the experimental data according to the random jitter characteristics of the centroid of laser spot caused by atmospheric turbulence, dividing the speckle images into different gray levels, and executing encoding and post-processing operations. Finally, experimental analysis for the extracted random sequence was carried out by the NIST test tools. Result shows that this sequence not only reachs the standard of true random numbers, its amount and randomness are also higher than those of the random sequence generated by the equal frame rate sampling method. In addition, the relationship between the normalized variance of laser speckle video and the optimal sampling interval was analyzed, which provided an important basis for further research.