Video anomaly detection using deep residual-spatiotemporal translation network

Video anomaly detection has gained significant attention in the current intelligent surveillance systems. We propose Deep Residual Spatiotemporal Translation Network (DR-STN), a novel unsupervised conditional Generative Adversarial (DR-cGAN) model with an Online Hard Negative Mining (OHNM) approach. The proposed DR-cGAN provides wider network to learn mapping from spatial temporal representations and enhance perceptual quality of synthesized images generator. During training, we take only frames normal events produce their corresponding dense optical flow. At testing time, compute reconstruction error local pixels between real flow then apply OHNM remove false-positive results. Finally, semantic region merging is introduced integrate intensities all individual abnormal objects into full output frame. DR-STN been extensively evaluated on publicly available benchmarks, including UCSD, UMN, CUHK Avenue, demonstrating superior results over other state-of-the-art methods both frame-level pixel-level evaluations. average Area Under Curve (AUC) value evaluation for three benchmarks 96.73%. improvement ratio AUC frame level 7.6%.