The technological core of the complex comprises algorithms based on deep convolutional neural networks and methods of global numerical optimization, applied for model identification and data integration from measurement devices. The algorithms provide accurate estimation of dense metric depth maps with automatic calibration using sparse RGB-D points, determined through visual-inertial odometry algorithms, following the single-view stereo matching concept. In addition to the camera, the hardware part of the complex includes a single-board computer with a processor and FPGA core, a lithium-polymer battery for autonomous power supply (replaceable with a power board for connection to an external DC power source), a touchscreen display for human-machine interaction (optional when used in robotic systems). All components are assembled in a single case.
ADVANTAGES OF THE DEVELOPMENT
An important advantage of the complex is the use of a standard video camera instead of specialized RGB-D sensors of foreign production, which are now difficult to acquire.
APPLICATION AREAS
The complex is implemented as a standalone solution and is intended for use in information and control systems of unmanned vehicles and mobile service robots to solve navigation and mapping (SLAM) tasks. It can also be used as standalone equipment for 3D scene reconstruction.