Drones are increasingly being used in outdoor environments, based on GPS as the key technology for localisation and mapping in autonomous flights. However, the technology for autonomous drone systems used for indoor localisation and mapping is still maturing, with most of the systems still under research and development.
Existing autonomous indoor systems rely on expensive sensors that are costly to operate for commercial businesses. As GPS is unavailable indoors, these sensors which include lasers, sonars or computer vision, are used as navigation tools to determine location. Besides being costly, the autonomous systems that rely on sensors are heavy in weight and/or have demanding processing requirements.
The cost-effective and lighter solution, the Semantic Depth Prediction System (SDPS) developed by the Singapore institute, uses a monocular camera to recreate a 3D scene by fusing object detection, semantic segmentation and depth estimation. The use of a monocular camera reduces the cost and size of the indoor drone, thus ensuring safer operations in confined indoor spaces.
Besides being used as a standalone system, this solution can be integrated with other available sensors to provide a more robust navigation solution for indoor operations.
The system's overall architecture is based on an open source machine learning library for research and production, and is extendable for future enhancements.
The navigation software is in the form of modular Application Programming Interfaces (APIs). This software can support any drones with specific flight computers that possess navigation functionalities.
The different components of the system include:
• On-board APIs for drones with on-board computers
• Off-board APIs for drones without on-board computers
• Separate APIs for raw images, object detection and depth estimation
• Separate APIs for training new image files
• Support for Nvidia-based GTX graphics cards and Cuda 9.0
The institute is keen to establish the following types of partnerships with MNEs or SMEs of all sizes:
i) Licensing agreement where the partner could license the technology and further develop it to introduce it to its customers.
ii) Commercial agreement with technical assistance where the institute would provide support for the Semantic Depth Prediction System.