Advanced Vision and Imaging

Computers can extract rich geometric information from ordinary images—measuring how far objects are, tracking how they move, and reconstructing the three-dimensional structure of a scene—by analyzing patterns of light across one or more camera views. These capabilities underpin autonomous vehicles, surgical robots, augmented reality, and any system that needs to understand physical space from visual input alone. Deep learning, particularly convolutional networks, has dramatically improved accuracy on tasks like depth estimation from a single image and dense optical flow, though models often struggle to generalize across environments they weren't trained on. Active research is pushing toward learning these representations without labeled ground truth, refining multi-view and light-field techniques for higher fidelity reconstruction, and tightening the feedback loop between visual perception and physical action in robotic systems.

Works

106,538

Total citations

1,811,710

Keywords

Stereo VisionDepth EstimationOptical FlowConvolutional NetworksMulti-View StereoUnsupervised Learning