How neural networks can make a drone your next best friend, by Tessie Hartjes and Iris Huijben
Download as PPTX, PDF1 like1,292 views
The document discusses the use of neural networks to enhance drone functionality, specifically for indoor navigation and task assistance. It introduces a solution based on visible light communication for precise indoor navigation and explores sensor fusion technologies to improve performance. Furthermore, applications such as object recognition and face recognition in various fields, including healthcare and safety, are also detailed.
How neural networks can make a drone your next best friend, by Tessie Hartjes and Iris Huijben
- 1. How neural networks can make a drone your next best friend Tessie Hartjes MASTER Electrical Engineering Electromechanics & power electronics BACHELOR Innovation Science Track: Energie Iris Huijben MASTER Electrical Engineering Track: Care and Cure - Signal Processing BACHELOR Pyschology & Technology Track: Robotica How neural networks can make a drone your next best friend Blue Jay Eindhoven 10-08-2016
- 3. www.bluejayeindhoven.nl The first domestic drone Showing the useful potential of drones 02
- 5. www.bluejayeindhoven.nl Biggest challenge 04 Accurate indoor navigation Flying indoor and between people = no room for error! - The drone will fly more stable when it ‘flies itself’ (autonomous) - But, existing solutions aren’t accurate enough
- 6. www.bluejayeindhoven.nl Our solution 04 Visible light communication: cm exact indoor navigation Online database - Every LED lamp has its own recognizable frequency - An online database holds the position of every lamp and is accessed by the drone - The camera of your smartphone registers the composed light and calculates your position with respect to the known configuration of the lamps Unique and precise solution within robotics domain but one problem: people only walk in the x- and y-plane need for sensor fusion
- 7. www.bluejayeindhoven.nl Our solution 04 Sensor fusion LiDAR Light detection and ranging sensor Altitude measurement based on timing differences of laser beams Optical flow Relative x and y displacement A camera that ‘sees’ relative movement through image displacement
- 8. www.bluejayeindhoven.nl Blue Jay as assistent 05 Object recognition Navigation Face recognition
- 9. www.bluejayeindhoven.nl Neural networks for flexibility 06
- 10. www.bluejayeindhoven.nl Training and testing 07 -45 -35 -25 -15 -5 5 15 25 35 45 -45 -35 -25 -15 -5 5 15 25 35 45 Output(degrees) Target (degrees) Target-output correspondence Yaw Pitch
- 12. www.bluejayeindhoven.nl Challenges & user cases 09 Object avoidance Health care Recognizing new people and objects Noise reduction Assistance / guidance Fire safety
Editor's Notes
- #11: In mijn bachelor eindproject ben ik vorig jaar bezig geweest met herkenning van draaiingshoek van een gezicht. Ik heb toen verschillende soorten neurale netwerken getraind op databases van foto’s van gezichten met meerdere draaiingshoeken en verlichtingshoeken. In deze grafiek zie je een van de betere resultaten. Hier heb ik ook gebruik gemaakt van de grootste trainingsdatabase met de meeste variantie. Nu aan jullie de vraag, hoeveel foto’s heb ik hiervoor gebruikt denken jullie?