Why Computer Vision is Critical for Commercial Success of Drone Industry
In 2015 drone industry received 80% of total investment in 5 years.
If you look at drone sales structure in 2015, you will see that it consists of custom drones worth hundreds of thousands of dollars (optimized for specific business cases) and mass production drones, which are bought by people to “fly, see the house from a height of 33 feet (10 meters), and eventually break down.”
Whereas the business of custom drones is highly marginal, it is unlikely to be scalable.
The business of mass production drones, in its turn, is, in fact, the business of selling robotic toys and is limited by 1-2 million sales per year in the toys market. Also, none of mass-market drones solves any of business issues really well.
The main hope for the growth of sales lies on a new generation of standard mass production drones for commercial use (goods delivery, construction, etc.).
Main requirements for such drones would be the following:
1. They should fly on big distances (requirement for long battery life, effective engines);
2. They should have control systems that would allow performing tasks with great accuracy without any human labour (pilots).
Existing control systems for drones based on GPS work with unacceptable accuracy (5-6 metres). What could this level of accuracy mean in practice? It could mean that, for example, a delivery drone can land not in your backyard, but in your neighbour’s backyard or even on a driveway.
This drawback could be improved by merging computer vision and analytics in one system for drone navigation that would make programming of autonomous drone flights highly accurate (up to a few centimetres).
Computer vision is crucial for any navigational system designed for such popular commercial use cases as goods delivery or pipeline inspection.
The importance of computer vision systems is also confirmed by Intel’s recent purchase of Ascending Technologies and some of its non-public market transactions.
Test results of technologies developed by Augmented Pixels show that it is possible to achieve very good accuracy even with standard sensors.
The implementation of additional subsystems such as Object Detection and Object Following allows us to solve commercially promising tasks in business with the help of drones and to make their autonomous use massive and ubiquitous!