Tracking your drone

Tracking allows the drone to interact with its environment, and depending on the effect you want to create – and whether you’re working outdoors or indoors, where GPS is unreliable – there are several options available.

Motion capture rigs

1. Vicon

Use: Indoor & Outdoor

The market leader when it comes to motion capture systems. The more expensive of the bunch. A full Bonita system suitable for UAV tracking could be had for under $20k. This brochure claims a full Bonita system could be had for as little as $10k, probably taking into account a smaller tracking volume than above.

See a video of the Vicon T-Series being used outdoors in full sunlight.

2. Optitrack

Use: Indoor & Outdoor

Using their online tool, a setup using 8 x of their Flex13 cameras (1280×1024 @ 120 fps) would come to £6774 to own and would cover an area of 7.3 x 7.3 x 2.5 metres and track 6 rigid bodies. Not sure if that rigid body limit is a software thing or a hardware thing. They have a Flex3 camera that is cheaper, but it’s 640×480 in resolution, so it would not be as accurate.

Used for the Lexus Amazing in motion – SWARM tracking

3. Visualeyez

Use: Indoor

Another Mocap system which could be interesting as their cameras come in a 3 x camera setup in a single unit, so it would be really easy to set up. Their claim is that a single one of these tripods (their smallest cheapest model) could do around 190 cubic metres of useful space. Multiple of these tripods can be combined for larger spaces. To summarise, a single tracker setup that can track 5 markers in a volume of about 6x6x4 metres for $30-35k. Easy-use system automatically calibrates multiple units; you can even move the sensors mid-capture and it’ll recover in a few seconds. Also their active marker system would not suffer from ID-swapping ever, where two trackers could otherwise swap IDs if they move close (Drone 1 becomes Drone 2 and vice versa).

GPS tracking

Use: Outdoor

Traditional GPS has an accuracy of about 3 metres and doesn’t update very often – 5hz is considered pretty good. Flight controllers rely on a bunch of different sensors to give them the ability to fly; generally they use inertial sensors that can give them an accurate idea of what just happened in the moment, how much the drone just rotated or how much it has seemed to move, but these values will drift over time as errors add up, so slower but absolute sensors (north is always north) like a compass and a GPS are used to correct the data from the inertial sensors.

Some OpenSource flight controllers have had support for RTK GPS for a while.

5. Piksi RTK GPS 

Use: Outdoor

A Kickstarter project originally, the Piksi RTK GPS was funded and is now shipping a development kit for $995. Ships with 2 radio modules, one stays on the ground as a reference point and one flies. It can get centimeter precision with a 50hz update.

6. Augmented Reality (AR) Markers

Use: Indoor & Outdoor

Having AR markers on the ground and tracking them with a camera on the drone is an option (or having the AR marker on the drone itself). These systems are based around patterns that are easy for the computer to find; once it has found one of the patterns it can figure out the cameras relation to the marker. The algorithms are really fast and if you give up the choice of specifying your own patterns, a library like Studierstube Tracker can look for 4096 different markers in an image with low CPU usage. One big drawback is that the markers are pretty visible and could be considered ugly. The allure of this idea is that the tracked space could be pretty damn big at just the cost of more paper to print markers on. It has the disadvantage that all the processing has to be made on the drone itself, so there is extra weight and power requirements. This puts some limit on the kind of camera that the drone could carry as well.

ETH in 2010 used a PixHawk flight controller using an onboard vision system (camera + CPU) and markers on the ground to navigate:

7. On board depth cameras

Use: Indoor & Outdoor

Conceptually similar to the idea of filling the space with easily (for the computer) recognisable AR markers, but using either a depth camera or color cameras with even more complicated algorithms to build up a 3D map of the environment for navigation. One disadvantage of this is that it’ll require a lot of processing to be done on the drone itself, so it’s not (in 2015) suitable for mini quads. The latest (at April 2015) and best example of this is this AscTec and intel collaboration that straps 6 Intel Realsense cameras onto an Asctec Firefly. It’s able to navigate through a forest at relatively high speeds.

8. Vision-based tracking systems

Use: Indoor & Outdoor

Simultaneous Localisation And Mapping (SLAM). There have been a number of attempts to use SLAM with single color cameras to build up a 3D map of the environment. Using a single camera and moving it around, the drone can figure out 3D information by locking onto points and seeing how they move in relation to the camera. An example by AscTec. “The tum_ardrone Robot Operating System (ROS) package allows to let the Parrot AR.Drone fly autonomously, using PTAM-based, visual navigation

9. Manual Piloting

Use: Indoor & Outdoor

As we’ve described in the section on INNOVATION, manual piloting is an effective way of piloting, and as of now likely to be among the cheaper of options.

Next page in toolkit: MLF on aerial filming in 360°