基于分层模型和鲁棒字典学习的背景差分炸点检测

Detection of Blast Point Based on Hierarchical Model of Background Subtraction via Robust Dictionary Learning

针对背景差分炸点检测方法中背景模型难以更新背景估计和运算复杂等问题,提出一种分层模型下基于鲁棒字典学习的背景差分炸点检测方法。为提高运算效率,该方法对图像帧建立3层金字塔分层模型,在每层将图像帧分割为互不重叠块,逐层以图像块为单位通过改进的鲁棒字典学习方法进行背景估计,与当前图像帧作背景差分实现炸点检测。采用炮弹炸点图像序列对所提出的方法进行了实验验证。实验结果表明,与现有炮弹炸点检测方法相比,该方法在准确率、误检率和鲁棒性方面均具有优越性能。

For the background estimation and huge computation problems of background model in the background subtraction blast point detection method,a blast point detection method is proposed based on a hierarchical model of background subtraction via robust dictionary learning. To improve the operation efficiency,a three-tier pyramid hierarchical model is established to divide each frame image into non-overlapping blocks. The blast points are detected from the subtraction between current frame image and image background estimation by using the improved robust dictionary learning method layer by layer. Experimental results on a large number of blast point image sequences show that the proposed method has superior performance in correct detection rate,false positive rate and robustness in comparison with the existing blast point detection method.