Special issue on smart cameras for real-time image and video processing

Smart cameras and smart camera networks are becoming fundamental pieces for intelligent environments, ranging from homes to buildings and cities, progressively being inserted into our lives. From smart surveillance systems composed of a multitude of smart camera nodes to small wearable cameras able to render a visual log of our daily activities, these cameras interact with each other and with a wealth of other smart devices, and of course the Internet. Thanks to the convergence of relevant technologies such as imaging sensors, embedded systems, video analytics and deep learning, this field is currently undergoing a rapid development. This technology opens the door to many application domains including video surveillance, autonomous driving, robots and drones, smart factory and industry 4.0, health monitoring and care, to name a few. Real-time image and video processing play a key role in most of these applications. The goal of this special issue is to present and highlight the latest developments in the smart camera community with the focus being on the real-time aspects of image and video processing. It attracted a large number of submissions from researchers active in the area of image processing and computer vision. After a careful peer review process, nine manuscripts were accepted for publication in this special issue, covering the topics of automatic camera calibration, image super-resolution, motion deblurring, multiple object tracking, face attribute recognition, human detection, image retrieval, and energy-efficient design of smart cameras. An overview of the nine accepted papers is provided below. Camera calibration is a necessary preliminary step for computer vision in the 3D world. In particular, extrinsic camera parameters need to be computed each time a camera changes its position, thus not allowing for fast and dynamic network re-configuration. Many works have tried to automate the process of camera calibration. However, there is still a lack of fully unsupervised and markerless approaches for camera calibration in literature. Garau et al. present an unsupervised framework for estimating the extrinsic parameters of a camera network, which leverages an optimized 3D human mesh recovery from a single image, and does not require the use of additional markers. Their framework can work with a single camera in real-time, allowing the user to add, re-position, or remove cameras from the network in a dynamic fashion. Ultra-High-Definition (UHD) cameras are commercially available, but still expensive for many applications. Superresolution (SR) techniques have been exploited to reconstruct high-resolution images or video without modifying the sensor architecture. In order to fix the artifacts observed in highly textured areas of images, Marin et al. propose a two-step SR method, called local adaptive spatial super resolution (LASSR). This method includes a machine learningbased texture analysis and a fast interpolation method that performs a pixel-by-pixel SR. They present an FPGA-based implementation of the LASSR method, which enables highquality 2–4 k super-resolution videos to be performed at 16 frames per second (fps), using only 13% of the FPGA capacity; this opens the way to reach more than 60 fps by executing several parallel instances of the LASSR code on the FPGA. Recently, deep convolutional neural networks (CNNs) have been employed to handle low-level vision problems, such as resolution reduction and motion blurring. However, most existing CNN-based approaches can either handle single degeneration each time or treat them jointly through * Caifeng Shan caifeng.shan@gmail.com