About this Research Topic
Computer vision’s historical goal is to mimic the human vision system to understand real-world scenes, both static and dynamic. One of the main actors that we are interested in capturing and understanding are humans. This task is particularly relevant in many scenarios. For example, in automotive detecting humans is fundamental to avoid pedestrians, in security and surveillance to track and identify people involved in illegal activities, in healthcare to provide remote assistance, in robotics to train machines to interact with humans, in social analysis to gather automatically who might have some potential sociological traits, or in gaming to offer new immersive and interactive interfaces based on detected human bodies.
The goal of this research topic is to advance the community towards different aspects of the automatic analysis of humans. Humans are complex in their nature, and so the pipelines to capture their behavior and interactions. Models to analyze them are typically composed of several interconnected steps: data acquisition, detection, tracking, and behavior inference. Each stage has its own challenges and criticalities, and its success contributes tremendously to the next blocks in the chain, hence, to reach the goals. Starting from images in which persons are detected and their poses estimated, moving to dynamic situations in which persons need to be tracked or re-identified across time and frames, all the way towards more complex goals in which action, activity, and social interactions of single and groups should be inferred. Each step is, indeed, of exponential complexity. Moreover, the analysis of humans in the real-world includes different environmental situations and acquisition conditions like static/dynamic scene, rigid and non-rigid deformations, multimodal types of data (audio-video-depth), and various camera viewpoints, such as fixed cameras for surveillance, active moving cameras, data from drones, and egocentric videos.
Thanks to deep neural networks, the last decade’s researches showed unbelievable improvement in the analysis of humans. Despite nowadays we can detect persons, recognize identities with faces, track in real-time multiple people, infer 2D and 3D poses, or segment part of the human bodies with unprecedented accuracy, such research tasks are far from being considered closed.
“Perceiving Humans” aims at inviting papers that advance the field of computer vision centered around capturing, perceiving, and understanding humans. We are interested in manuscripts that push forward theoretical and practical aspects of methods that automatically sense humans or have critical roles in such tasks. Topics of interest include but are not limited to: Detection, Tracking, Re-Identification, Social signal processing, Action/Activity recognition, Egocentric vision, Video surveillance, Human pose estimation, Shape estimation, Forecasting, Human-Robot interaction and cooperation. Moreover, applications on the frontiers, such as performing arts blending computer vision, are more than welcome.
The types of accepted manuscripts are Original Research, Systematic Review, Methods, Hypothesis and Theory, Data Reports, and Perspective.
Keywords: humans understanding, action/activity recognition, tracking, pose estimation, ego-centric vision, social signal processing.
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.