针对基于传统混合高斯模型算法对光照突变敏感,以及运动物体速度缓慢时产生的"重影"现象,根据深海自主观测的应用需求提出了一种三帧差法结合动态自适应学习率的改进混合高斯模型算法。利用三帧差法快速定位出视频帧的背景、前景及干扰区,对不同区域给予不同的学习率和更新策略,减小了运算量和提高了算法的收敛速率。实验结果表明,该算法能有效地处理深海环境中海流扰动、浮游物干扰以及光线突变带来的影响,具有良好的实时性和鲁棒性,在深渊着陆器的实际工程应用中能够准确地检测出运动目标。 ; Aiming at engineering application in deep-sea, this paper presents an improved algorithm of Gaussian mixture model based on three-frame differencing blocking model and adaptive learning rate for the problems of difficulties of the illumination change and the phenomenon of ghosting caused by slow speed of moving targets. The image of each frame is divided into background, foreground and interference area through three-frame differencing, different area is given corresponding learning rate and update strategy, thus the convergence rate get improved and largely decreasing the calculation. Experimental results show the algorithm can effectively solve the problems caused by current disturbance, seston moving and light changing, which has good real-time performance and robustness, and can detect moving targets accurately in the application on deep-sea lander.