A couple of videos have shown up on YouTube showing quadcopter control using image recognition on the Jetson TK1 and Tegra K1 based systems. The first is from Kazuya Sato which shows a PG400 enRoute quadcopter being controlled from a Jetson TK1. Looky here:
What’s going on here?
I couldn’t read the description, but what appears to be happening is that the Jetson is connected to an enRoute D8T-2.4G radio transceiver on a micro controller through USB. The micro controller appears to be an Arduino, but it’s hard to tell from the video, as the transceiver is piggy backed on top of the micro controller. This hardware allows the Jetson to be able to send control commands to the quadcopter.
A Logicool Qcam Orbit AF webcam is also connected to the Jetson, and is pointed towards the quadcopter from a tripod. Two visual markers, a red ball and a yellow ball are placed on the quadcopter.
The user denotes on the video image displayed on the Jetson where they would like to place the quadcopter in the air. By optically tracking the markers, the Jetson sends commands to the quadcopter telling it to navigate to the desired position. The tracked markers, combined with IMU information from the quadcopter, allow for a feedback control system which is used to position the drone precisely where the user desires. This also allows for the system to potentially take into account and correct for external forces, such as wind, that a small drone like this typically encounters.
The second video is from a blog entry by Kai Yan on the diydrones.com. Looky here:
Kai Yan has a diydrones blog entry describing in detail what’s going on: “Controling a copter by image recognition“. A summary is that a quadcopter has a red circle on it, and a NVIDIA Shield Tablet (with a Tegra K1 processor) tracks the red circle and issues commands to the quadcopter to keep the drone in the center of the field of view of the tablet camera. enRoute plans to make the code available through their Github account when a few more features are added.
One of the ways these systems can be used is for situations where GPS is weak and traditional autonomous navigation needs to be augmented, such as under bridges or in cities where large buildings obstruct GPS signals.