Articles tagged with Computer Vision
Researchers have developed novel approaches to resolve low-level vision in videos caused by rain and night-time conditions, as well as improve 3D human pose estimation in videos. These techniques can be used to enhance the quality of night-time videos and rain videos, addressing visibility issues during these environmental factors.
Researchers from Skoltech have developed a new augmentation technique called MixChannel to help train computer vision algorithms with limited data. This approach outperformed state-of-the-art solutions in testing with three neural networks and can be combined with other methods for even more training data.
Researchers from UTSA, UCF, AFRL, and SRI International have developed a new method that improves how artificial intelligence learns to see. By adding noise to multiple layers of a neural network, the team creates more robust representations of images recognized by AI, leading to better explanations for AI decisions.
A Kanazawa University researcher has developed a method to speed up non-rigid point set registration, a fundamental problem in computing with extensive applications in autonomous driving, medical imaging, and robotic manipulation. The proposed technique reduces computing time for large point sets, outperforming state-of-the-art approac...
Researchers have detected bias in face recognition algorithms, with higher false positive rates for females with dark skin tone and males with light skin tone. Top winning solutions exceeded 99.9% accuracy, but the analysis of top 10 teams showed that overall accuracy is not enough when building fair face recognition methods.
UT Arlington computer scientists develop a deep learning method to generate synthetic objects for robot training, overcoming the need for manual capture of images from human-centric perspectives. The technique uses generative adversarial networks (GANs) to create photorealistic full scenes and dense colored point clouds with fine details.
Skoltech researchers use chemical sensors and computer vision to monitor grilled chicken doneness, promising automation in kitchen quality control. The system accurately identifies undercooked, well-cooked, and overcooked chicken.
A team of researchers from Duke University has developed a method to make neural networks more transparent and interpretable. By modifying the reasoning process behind predictions, it is possible to better understand how these complex models work. The approach involves replacing standard parts of a neural network with new ones that con...
A new computer vision app developed by University of Cambridge engineers allows easier monitoring of blood glucose levels in people with diabetes. The app uses a smartphone camera to read glucose meter data, eliminating the need for manual input or internet connectivity.
A Cornell-led team is using robots with computer vision to optimize apple yields by controlling fruit numbers per tree. This can increase total crop values by $7,000 per acre, helping growers meet market demand for fruit size and quality.
A University of Georgia researcher used computer vision to analyze thousands of images from over 100 Instagram accounts of United States politicians, discovering that posts featuring politicians' faces in non-political settings attract more likes and comments. The study found that images with only the politician's face or in personal s...
Feitian Zhang and Pei Dong received a $15,000 grant to purchase a remotely operated vehicle with GPS for monitoring microelectronic sensors in aquatic environments. The funding will support their research in aquatic environmental monitoring until August 2021.
A new study by Johns Hopkins University researchers found that the brain detects 3D shape fragments in the early stages of object vision, a strategy also used in artificial intelligence networks. This discovery may hold future opportunities to leverage correlations between natural and artificial intelligence.
Researchers from the University of Bristol and Manchester have developed cameras that can learn and process visual information in real-time, eliminating the need to record and transmit images. This breakthrough enables intelligent machines to perceive the world more efficiently and securely.
Researchers at Princeton University developed a tool to uncover potential biases in visual data sets, such as stereotypical images and underrepresentation. The tool, REVISE, uses statistical methods to inspect data sets for object-based, gender-based, and geography-based biases.
Researchers at Cornell University developed a method to create maneuverable 3D images showing changes in appearance over time using deep learning and tens of thousands of publicly available tourist photos. The tool, called Deep Multiplane Images, allows users to explore scenes from different viewpoints and time frames.
Researchers have developed an efficient method to estimate camera movement, reducing the number of hypotheses generated from up to five to one. The new approach allows for real-time execution of pose estimation, with a complete algorithm taking only 29 milliseconds per frame.
Korean researchers from ETRI win first and second places in the Challenge on Learned Image Compression (CLIC) with innovative AI-based video compression techniques. The team aims to optimize video compression rate and quality using multiple source technologies.
Researchers at Carnegie Mellon University have developed a system that combines iPhone videos to create 4D visualizations, allowing viewers to watch action from various angles. The method uses convolutional neural nets and can be applied to a wide range of scenes, including those shot independently from different vantage points.
Researchers developed a software framework that incorporates computer vision and uncertainty into AI for robotic prosthetics, allowing users to walk safely on various terrains. The framework uses robust AI algorithms to predict terrain type, quantify uncertainty, and adjust behavior accordingly.
Researchers at SLAC National Accelerator Laboratory used computer vision and X-ray tomography data to understand how nickel-manganese-cobalt cathodes degrade over time. They found that particles detaching from the carbon matrix contribute significantly to battery decline, contradicting previous assumptions about making smaller particle...
A new algorithm for segmenting biological objects in complex images has been developed by Skoltech researchers. The method uses a two-step neural network training algorithm that can learn from small datasets and achieve high accuracy in isolating individual cells, organisms, and parts of plants.
A new computer model developed by MIT cognitive scientists can quickly generate a detailed scene description from an image, similar to the brain's ability. The model, known as efficient inverse graphics (EIG), reverses the steps used in computer graphics programs to generate images, allowing it to infer underlying features of a scene. ...
A team of researchers from Princeton and Stanford University has developed methods to obtain fairer data sets containing images of people. They propose improvements to ImageNet, removing non-visual concepts and offensive categories, such as racial and sexual characterizations.
Researchers found a brain circuit that enables fruit flies to see in color, similar to the human capacity for color vision. The study sheds light on the transmission of information from the eye to the brain and could inspire future technologies for those with vision impairments.
A team led by John Tsotsos disproved a long-standing theory of how the human vision system processes images. The study found that salience is not needed for quickly deciding what an image depicts and that current AI algorithms fall short in matching human performance.
A team of researchers is developing a data-driven computer model to diagnose and treat strabismus, a prevalent condition affecting 18 million people in the US. The model will use clinical information from MRI scans, surgical procedures, and patient outcomes to improve treatment options.
The University of Pittsburgh has received a $6 million grant from the Richard King Mellon Foundation to support the development of a cortical vision research program. The program aims to understand how the eye and brain work together to restore vision, using cutting-edge technologies such as brain computer interfaces and optogenetics.
A robot developed by the University of Cambridge has successfully harvested iceberg lettuce in various field conditions, demonstrating potential for expanding robotics in agriculture. The 'Vegebot' uses machine learning to identify healthy lettuces and cut them without crushing, reducing physical demands on manual harvesting.
Researchers at Lancaster University created a series of games that require players to use their peripheral vision, resulting in significant improvements in object recognition. The study found that even just one gaming session led to lasting improvements in peripheral awareness, suggesting potential applications in team sports and hazar...
Computer scientists at the University of Washington have developed an algorithm, Photo Wake-Up, that can animate people from 2D photos. The system uses a combination of 3D template matching and texture pasting to create realistic animations in three dimensions using augmented reality tools.
Researchers have created a new dataset called BOLD5000, which comprises brain scans of four volunteers viewing 5,000 images. This dataset allows cognitive neuroscientists to leverage deep learning models that have improved artificial vision systems.
A novel system developed at MIT uses RFID tags to help robots home in on moving objects with unprecedented speed and accuracy. The system, called TurboTrack, can locate tagged objects within 7.5 milliseconds, on average, and with an error of less than a centimeter.
Researchers have developed a computer vision system that uses a brain-inspired approach to learn and identify objects in real-world images. The system is trained on a vast amount of data from the internet, allowing it to build a detailed model of objects without external guidance.
Researchers at RIT are developing an advanced visual tracking system using deep learning and artificial intelligence to refine object location and movement. The system has potential applications in autonomous navigation, drones, traffic monitoring, safety, security, disaster response, and human-computer interaction.
The UC3M and Álava Engineers have created a professorship to encourage research in computer vision, focusing on image capture and analysis. The joint project will develop applications for offline and online image processing using various technologies.
A new tool developed at Princeton University streamlines the creation of computer-animated images by automatically separating repeating objects into layers. The tool allows users to manually select and draw motion lines, which are then used to animate similar elements in a believable manner.
The SWEEPER robot, developed by an international research consortium, can harvest ripe fruit in 24 seconds with a success rate of 62 percent. Additional research is needed to increase work speed and reach higher harvest success rates.
Researchers at The University of Tokyo developed a computational tool that can learn from headcam footage to predict where the user's focus will next be targeted. This approach combines visual saliency with gaze prediction, achieving better results than existing methods.
Computer vision algorithms have made significant progress in tasks such as object identification and categorization. However, they struggle with determining whether two objects in an image are the same or different. Researchers at Brown University found that this limitation stems from the inability of these algorithms to individuate ob...
Researchers developed a new method to train computers to better recognize objects in the real world by using virtual reality. A virtual dataset called ParallelEye was created, allowing for diverse and realistic images of various scenes, which significantly improved performance on object detection tasks.
Researchers at Newcastle University have discovered a new form of 3D vision in praying mantises that works differently from previously known forms. This unique vision system allows mantises to detect movement and distance without detailed image matching, making it robust and efficient for processing.
Researchers at Duke University Medical Center found players with higher scores on vision and motor tasks completed on large touch-screen machines had better on-base percentages, more walks, and fewer strikeouts. High scores in perception-span task were associated with an increased ability to get on base.
Human brains tend to miss objects that are mis-scaled, even when they're in view. Researchers found this phenomenon in eye-tracking studies, but not in computer vision algorithms like deep neural networks. This study aims to better understand human visual search strategies and improve computer vision.
Researchers have developed a web app capable of producing 3D facial reconstruction from a single 2D image. The technique, using Convolutional Neural Networks, allows for arbitrary facial poses and expressions, with over 400,000 users already trying it out.
A team of researchers from the University of Texas at Arlington is working with Macnica Americas to evaluate existing deep learning methods for face detection and facial recognition. The collaboration aims to improve performance and reduce computational load to make the technology more available to customers.
Researchers from UC San Diego showcase self-folding robots, robotic endoscopes, and improved computer vision techniques to enhance human-robot collaboration. The conference focuses on developing friendly robots that can work effectively with humans in various domains.
Researchers developed a new way to assess and predict facial expressions of movie-goers using factorized variational autoencoders (FVAEs). The method demonstrates a surprising ability to reliably predict viewers' facial expressions for the remainder of the movie after just a few minutes of observation.
Researchers at the University of Surrey have developed a new approach to deliver immersive audio experiences by utilizing all available devices in a living room, such as laptops and wireless mini-speakers. The 'Media Device Orchestration' concept enables users to enjoy spatial audio in a more immersive and multi-layered way.
Researchers at Disney Research and UC Davis have developed a method for computer vision programs to understand spatial relationships in images based on caption sentence structure. This approach enables accurate visual localizations for language inputs, outperforming baseline systems that do not consider natural language structure.
Researchers have developed a method for designing energy-efficient neural networks, reducing power consumption by up to 73% compared to standard implementations. The new approach uses an analytic tool to evaluate and prune low-weight connections, resulting in more efficient networks with fewer connections.
A computer vision system analyzed street-level photos to gauge neighborhood safety and predict urban change. The study found that the density of highly educated residents, rather than income or ethnic composition, predicts revitalization in five American cities.
Research suggests dressmakers have superior stereovision, improving visual acuity by 43%. This sharpens their ability to thread needles and navigate 3D spaces.
The GelSight sensor uses physical contact to provide a detailed 3-D map of an object's surface, enabling robots to judge the hardness of surfaces they touch. Researchers also use it to enable robots to manipulate smaller objects than previously possible.
A new method developed by Deva Ramanan and Peiyun Hu reduces error rates for detecting tiny faces in images by a factor of two, resulting in 81% accuracy. The approach leverages context, including body shapes and crowd compositions, to improve object detection.
Researchers at Disney Research developed an AI-based system that can automatically learn the association between images and sounds, with applications in film sound effects and aiding visually impaired individuals. The system uses video data to filter out uncorrelated sounds and learns which sounds are associated with an image.
A $450,000 grant will fund a collaboration between Indiana University and the US Navy to develop new methods for inspecting microelectronic components used in critical military systems. Computer vision technology will be applied to improve the integrity of electronic circuitry, reducing defects and ensuring equipment reliability.
Researchers from Disney Research and Fudan University developed a new approach that enables computers to recognize events in videos, including categories of events they've never seen before. The software associates visual elements with each type of event and can learn from new examples to improve its accuracy.
Computer vision systems can now learn to recognize objects they have never seen before by analyzing word use and contextualization, reducing the need for thousands of labeled images. This new learning paradigm, called semi-supervised vocabulary-informed learning, was developed by Disney Researchers using a large dataset of English words.
Researchers at North Carolina State University developed a new image segmentation technique that improves object identification and separation in images. The technique, called Consensus-Based Image Segmentation via Topological Persistence, aggregates data from multiple algorithms to create a new version of the image.