针对传统混合高斯模型使用固定学习速率所带来的问题,提出了一种改进的运动目标检测算法.该算法采用自适应的学习速率调整策略,在背景建模初期,采用较大的学习速率加快初始背景的建模,使得模型能更快地适应背景的变化；背景形成以后,根据目标运动的快慢动态调整学习速率,从而能够及时更新背景,消除运动目标的残留和拖影；最后利用基于HSV颜色空间的阴影检测算法消除运动阴影.实验结果表明,改进算法优于传统混合高斯模型,可以更准确地检测出运动目标,更好地消除阴影,并具有较好的自适应性和稳健性.%To solve the problem of traditional Gaussian mixture model using a fixed learning rate, an improved algorithm of moving objects detection is proposed. The algorithm adopts adaptive adjustment strategy for learning rate. Early in the background modeling, a larger learning rate is adopted to accelerate the background modeling, so as to make for the model adapting the change of backgrounds faster. After background is formed, the learning rate is dynamically adjusted by the moving speed of objects, which can update the background ttimely, and eliminate the residual objects and smear. Finally, moving shadows are eliminated by the shadow detecting algorithm based on HSV color space. The experimental results show that the improved algorithm is better than traditional Gaussian mixture model. It can more accurately detect moving targets and eliminate shadows preferably, and has better adaptability and robustness.
展开▼
