基于改进混合高斯模型的自适应运动车辆检测算法

摘要

There are often the cases in road video surveillance systems that the vehicles are slowly moving or in short stay.In view of the problems that the background subtraction method of traditional Gaussian mixture model is sensitive to abrupt changes in environment and has information loss on slow moving target,we propose an improved adaptive vehicle detection algorithm.First,in order to restrain the foreground of slow movement to be trained to the background,the present pixel-values are classified before updating the parameters,and the models are set different replacement rates according to classification results.Secondly,for removing the interference of environmental changes,a metric factor that tracks environmental changes is introduced to realise the adaptive switch between the background subtraction and the inter-frame difference algorithm when abrupt environmental change occurs.Finally the more accurate object is gotten by ecological filtering.Experiments show that this algorithm can get better detection effect for moving vehicles in daytime real-time traffic video.%道路视频监控中经常存在车辆缓慢运动或短暂停留的情况。针对传统混合高斯模型背景减除法对环境突变敏感和对缓慢运动目标丢失信息的问题，提出一种改进的自适应车辆检测方法。首先，在参数更新前对像素值分类并根据分类结果设置模型更新率，抑制缓慢运动前景被训练成背景；引入一个跟踪环境变化的度量因子，当环境突变时实现背景减除和帧差法的自适应切换，滤除环境变化的干扰；最后通过生态学滤波得到准确的运动目标。实验表明，该算法对白天实时路况视频中的运动车辆具有较好的检测效果。