An algorithm utilizing RGB color features and scale invariant local ternary patterns is presented for sol-ving the difficulty of detecting moving targets in complex scenes. The time-domain median method was adopted to estimate background image and initialize background model quickly. By fusing similarity measures of color and tex-ture features, a background probability network was obtained. The application of lateral inhibition filtering improved the contrast, the foreground and background pixels were classified, and shadow detection worked for the foreground pixels. The shadow pixels were classified as background pixels but not used for the model update. The performance of the proposed algorithm was compared with the other three algorithms in the change detection database. The pro-posed method can accurately handle scenes containing moving backgrounds, shadows, and camera jitter, with ac-ceptable real-time performance.%针对复杂场景中运动目标检测这一难题,提出利用RGB颜色特征和尺度不变局部三元模式的运动目标检测算法. 利用时域中值法得到估算背景图像并快速初始化背景模型. 通过颜色特征、纹理特征相似性度量,融合得出背景概率网络,通过侧抑制滤波提高对比度分类出前景与背景像素,前景像素进一步进行阴影检测,将阴影点归为背景点,但不用于模型更新. 将算法与GMM、SC-SOBS、SUBSENS算法在变化检测数据库中进行对比验证. 实验表明,新算法在满足实时性的基础上,对动态背景,阴影和相机抖动等有一定的鲁棒性.
展开▼
