Real-Time Stopped Object Detection by Neural Dual Background Modeling

Moving object detection is a relevant step for many computer vision applications, and specifically for real-time color video surveillance systems, where processing time is a challenging issue. We adopt a dual background approach for detecting moving objects and discriminating those that have stopped, based on a neural model capable of learning from past experience and efficiently detecting such objects against scene variations. We propose a GPGPU approach allowing real-time results, by using a mapping of neurons on a 2D flat grid on NVIDIA CUDA. Several experiments show parallel perfomance and how our approach outperforms with respect to OpenMP implementation.