Automatic monitoring systems, self-driving cars, facial detection, wellness, and social distancing tools are all examples of where computer vision technology is being employed. Users need accurate and trustworthy visual data to properly benefit from video analytics solutions, however, video network performance is frequently impacted by environmental variables such as rain, night-time circumstances, or overcrowding (where there are multiple images of people overlapping with each other in a scene). A group of researchers headed by Yale-NUS College Associate Director of Science (Computer Science) Robby Tan, who is from the National University of Singapore’s (NUS) Faculty of Engineering, has formed innovative methods to remedy the issues of vision problems in videos heavy rainfall and night-time situations, as well as increases the reliability of the videos using machine learning and deep learning.
Getting About in Rain and at Night When Visibility Is a Problem
Low lighting and man-made specular lightings such as dazzle, flare, and strobing lights impact night-time photos, whereas rain images are influenced by rain streaks or build-up (or rain veiling effect). “To function properly, many computer vision systems, such as automated surveillance and self-driving automobiles, require a good view of the input footage. Self-driving vehicles, for example, cannot function reliably in heavy rain, and CCTV automated surveillance systems frequently fail at night, especially if the settings are gloomy or there is substantial glare or floodlights “Assoc Prof Tan said.
The primary sensor, i.e. numerous persons within the same frame, has an impact on human object detection, especially when individuals are conversing nearby or appear to be merging in the binocular video.
Assoc Prof Tan and his team used deep learning algorithms to improve the quality of night-time and rain-related films in two independent investigations. In the first experiment, they increased the luminance while reducing noise and light effects (glare, glow, and floodlights) to provide crisp nighttime pictures. This is a novel approach that solves the issue of clarity in night-time photos and movies when glare cannot be avoided. On contrary, current state-of-the-art techniques are incapable of dealing with glare.
3D Human Pose Prediction: Addressing Inaccuracies in Videos Due to Overlapping, Numerous Persons.
Assoc Prof Tan also shared his team’s study on 3D human posture prediction at the CVPR meeting, which has applications in surveillance cameras, video gaming, and sports reporting. Assoc Prof Tan also shared his team’s study on 3D human posture prediction at the CVPR meeting, which has applications in surveillance cameras, video gaming, and sports reporting.
Researchers and engineers have been progressively focusing on 3D multi-person posture estimation from a monocular video (video captured from a single camera) in recent times. Monocular movies, rather than utilizing several cameras to capture footage from various locations, offer greater flexibility since they may be captured with a single, conventional camera — even a mobile phone camera.
In this related experiment, the scientists combine two known approaches to estimate 3D human postures from a video, ie a top-down technique and a bottom-up one. The novel method may generate more consistent posture estimation in multi-person scenarios and manage the distance between persons (or scale changes) more effectively by integrating the two ways.