基于FPGA的混合高斯背景建模实现

Implementation of real-time Gaussian mixture models based on FPGA

为了解决PC机上高清视频运动目标检测的实时性瓶颈问题,设计了一种基于FPGA的运动目标检测系统。系统采用基于自适应混合高斯背景模型的背景差分法,对环境扰动具有很好的适应性。本设计应用于1 280×1 024高清视频的运动目标检测,针对硬件实现的特点,对OpenCV混合高斯背景模型算法进行改进和适当的参数定点化,设计了适合FPGA实现的结构,在Altera Stratix IV开发平台上成功实现了高清视频的背景建模和目标检测,试验表明效果良好,满足实时检测的要求。

In order to overcome the bottleneck of real-time in moving targets detection of HD video based on PC, this paper designs a moving targets detection system based on FPGA. The system adopts background subtraction based on adaptive Gaussian mixture models, and has good adaptability in non-stationary background scenarios. The design applies to moving target detection of 1 280×1024 HD video. According to the characteristics of hardware, we improve Gaussian mixture background models algorithm in OpenCV, formulate fixed-point scheme, and design the circuit structure which is suitable for FPGA. We successfully realize the background modeling and target detection on Altera Stratix IV development platform, good performances have been demonstrated in the experiment, and we meet real-time demand excellently.