摘要
鉴于目前基于图像的运动目标检测系统对目标场景光照条件非常敏感,文中提出一种基于Kinect深度数据曲率的单高斯模型运动目标检测方法,增强了系统对场景采集误差的鲁棒性。首先对深度数据进行中值滤波,利用单高斯模型对目标区域深度数据进行建模;在对目标场景实时采集数据与背景参数进行高斯概率门限值判别后,经过形态学滤波,达到了运动目标检测的目的。同时利用实时更新背景参数的方法提高模型适应场景变化的能力,并通过实验取得了良好的检测效果。
The moving target detection systems based on image processing are sensitive to the light conditions of target scene. Present a new single Gauss model moving target detection method based on Kinect depth data,which can enhance robustness of the system in scene collection error. First,median filtering is used to process depth data. Then the single Gauss model is established for the depth data of target scene. The Gauss probability threshold is used to discriminate the collecting data of target scene and background parameters. Moving tar-get detection is achieved after morphological filtering. And real-time updating background parameters is a good way to make the model a-dapt to the change of scene. At last,experiments are performed. The proposed method has achieved well detection results.
出处
《计算机技术与发展》
2015年第9期27-30,共4页
Computer Technology and Development
基金
国家自然科学基金资助项目(61005015)
国家第三批博士后特别基金(201003280)
上海市青年教师培育计划
