A passive visual dense monocular deformable SLAM code. (in development)
Intended to run on any GPU on any OS. It uses OpenCV to load images, but all processing in directly in OpenCL.
This project is implementing a passive visual dense monocular deformable SLAM code.
Passive : no active depth sensor required
Visual: based on rgb video images as the input
Dense: reconstructs 3D depth at every pixel
Monocular: requires only one camera
Deformable: handles scenes where objects move and change shape - e.g. people, animals, soft materials and fluids
SLAM: simultaneous location and mapping aka structure from motion.
The road map for the project includes (i) auto-calibration of the camera and algorithm parameters, (ii) integrating single image shape, illumination and reflectance modelling.
Use cases:
1) Visual perception for robotic manipulation and autonomous vehicles
2) Computer Graphics models from live action video for reality capture and VFX.
View full history Series and milestones
trunk series is the current focus of development.