张 恒, 胡文龙, 丁赤飚(中国科学院空间信息处理与应用系统重点实验室, 北京 100190)
Adaptive Learning Gaussian Mixture Models for Video Target Detection
ZHANG Heng, HU Wenlong, DING Chibiao(Key Laboratory of Spatial Information Processing and Application System Technology, Chinese Academy of Sciences, Beijing 100190)
Background subtraction is a widely used method for video object detection and its performance is dependent on the quality of background model. In this paper, an algorithm for video target detection based on adaptive learning GMM was proposed by defining an efficiency factor between pixel samples and their background models. The accumulation of efficiency factor(AEF) shows how well the models can represent the background and was used to adjust the learning-rate dynamically. At the same time, how to update the models was dependent on the changes of the background after the foreground image analysis. The performance and robustness of the algorithm has been verified experimentally.