To See in the Dark: N2DGAN for Background Modeling in Nighttime Scene

Due to the deteriorated conditions of illumination lack and uneven lighting, performance traditional background modeling methods is greatly limited for surveillance nighttime video. To make under scene performs as well in daytime condition, we put forward a promising generation-based framework foreground surveillance. With pre-specified reference image frame, GAN based generation model, called N2DGAN, trained transfer each frame video virtual with same except part. Specifically, balance preservation object(s) generating image, presented two-pathway which global local sub-networks were combined spatial temporal consistency constraints. For sequence generated images, multi-scale Bayes model was further proposed characterize pertinently variation background. We manually labeled ground truth on collected nightime datasets evaluation. The impressive results illustrated both main paper supplementary show effectiveness our approach.