Performance comparison of human silhouette extraction from videos by semantic segmentation with ResNet-101 compared to statistical background modelling.