Computer vision technology is becoming widely popular in several sectors such as healthcare, self-driving cars, social distancing tools, automatic surveillance systems, and facial recognition. To enjoy the complete benefits offered by video analytics applications, users need accurate and dependable visual information. However, the quality of video data received is at times affected by environmental factors such as crowds, rain, or night-time conditions.
To tackle the problem constantly faced by users, researchers have now developed few innovative techniques that could enhance the low-level visions caused due to rain and night time conditions. The new study might boost the Computer Vision Market as it not only removes the existing banes surrounding the technology but also enhances the accuracy of 3D human pose estimation present in video data.
Two studies presented on the same topic demonstrated that deep learning algorithms could be used to improve the quality of night time and rain videos, respectively. In one study, the researchers showed the way in which the brightness in the video could be boosted while noise and light effects (glow, floodlights and glare) could be suppressed, thus, leading to a clear night-time image. The novel technique successfully addressed the problems encountered in providing night-time videos and images in the presence of glare, something that current state-of-the-art methods could not handle.
In another study, problems faced by rain were took-on. Tropical countries like Singapore have rain year-round, and the rain veiling effects can lead to degraded videos. To tackle the challenge, the team introduced a method that makes use of frame alignment. The technique would enable procurement of better visual information without being held back by rain streaks which have been the cause of the bad quality of image many times. Thereafter, the team employed a moving camera to enable depth estimation resulting in the removal of the rain veiling effect. Distinct to currently available approaches that invest in removing rain streaks, the new method instead removes rain streaks as well as rain veiling effect simultaneously.
In the third study, researchers expected 3D human poses with the help of a video by merging two already existing techniques, namely, the bottom-up and top-down approaches. They combined them to form the new method, which gives an output of much more reliable pose estimation in a setting with several humans. In addition, it also handles the distance between individuals much more robustly. After the success of these studies, researchers have revealed that their next step would be to find ways to protect the privacy of these videos.