Foreground detection is an important theme in video surveillance. Conventional background modeling approaches build sophisticated temporal statistical model to detect foreground based on low-level features, while modern semantic/instance segmentation approaches generate high-level foreground annotation, but ignore the temporal relevance among consecutive frames. In this paper, we propose a Spatio-Temporal Attention Model (STAM) for cross-scene foreground detection. To fill the semantic gap between low and high level features, appearance and optical flow features are synthesized by attention modules via the feature learning procedure. Experimental results on CDnet 2014 benchmarks validate it and outperformed many state-of-the-art methods in seven evaluation metrics. With the attention modules and optical flow, its F-measure increased 9% and 6% respectively. The model without any tuning showed its cross-scene generalization on Wallflower and PETS datasets. The processing speed was 10.8 fps with the frame size 256 by 256.

Terahertz imaging (frequency between 0.1 to 10 THz) is a modern technique for public security check. Due to poor imaging quality, traditional machine vision methods often fail to detect concealed weapons in Terahertz samples, while modern instance segmentation approaches have complex multiple-stage concatenation and often hunger for massive and accurate training data. In this work, we realize a novel Conditional Generative Adversarial Nets (CGANs), named as Mask-CGANs to segment weapons in such a challenging imaging quality. The Mask-Generator network employs a "selected-connection U-Net" to restrain false alarms and speed up training convergence. The loss function takes reconstruction errors and sparse priors into consideration to preserve precise segmentation. Such a learning architecture works well with a small training dataset. Experiments show that the proposed model outperforms CGANs (more than 16-32 % in Recall, Precision and Accuracy) and Mask-RCNN (more than 3-6 %). Moreover, its testing speed (69.7 FPS ) is fast enough to be implemented in a real-time security check system, which is 44 times faster than Mask-RCNN.In the experiments for mammographic mass segmentation on INBreast dataset, the Dice index of the proposed method is 91.29, surpasses the-state-of-the-art

No abstract

No abstract