Spatiotemporal saliency model for small moving object detection in infrared videos

Abstract In this paper, a novel spatiotemporal saliency model based on three-dimensional Difference-of-Gaussians filters is proposed for small moving object detection in infrared videos. First, instead of utilizing the spatial Difference-of-Gaussians (DoG) filter which has been used to build saliency model for static images, we propose to extend the spatial DoG filter to construct three-dimensional (3D) Difference-of-Gaussians filters for measuring the center-surround difference in the spatiotemporal receptive field. Second, an effective spatiotemporal saliency model is generated based on these filters. This model provides a good basis for accurate and robust infrared small moving object detection. Experimental results show that the proposed saliency model consistently outperforms state-of-the-art saliency models for infrared moving object detection under various complex backgrounds.