My name is Matt, a grad student at Colorado state university studying mechanical engineering. I am going to attempt to build an autonomous quadcopter and looking for a contact who can answer some questions. I would like the UAV to be able to track a specified object either ground to air or air to air or both. I have an idea of what sensors and maybe what type of frame and operating system I would need but I have more questions than can be written in a single post. Is some one willing to have a conversion about this subject? I would greatly appreciate some guidance and direction.
Thank you very much,
you might get some good responses on here, but I’d suggest also dropping in to https://gitter.im/ArduPilot/VisionProjects
Detecting, identifying and tracking/following an object is a really challenging project - good luck!
FYI, if you haven’t already found it, Randy’s Red Balloon Finder is a few years old now, but still a good place to start your research for this project I think. http://ardupilot.org/dev/docs/odroid-via-mavlink.html#Red_Balloon_Finder
Thanks james_pattison for the response. Can you tell me why tracking and identifying by an autonomous aircraft if a tough task? Do we not have sensors available to detect this sort of thing or is it just to expensive to investigate? I do know we have sensors available to track and follow particular colors and sensors that can be specified to stay a particular distance away from an object but I’m unsure if distance sensors can work vertically or aerially (it’s part of my investigation). I am also looking to build this UAV from ground up using either a kit or by building from an assortment of parts and 3D printing particular pieces I need. Can you give input as to how I should get the quadcopter build going? I am also open to other types of UAV’s with the capability of being able to hover and fly backwards.
It isn’t just sensors: it’s integrating the right sensor mix with the right computer and software into the flight control system, all in a package that is power efficient, small and lightweight. It’s definitely achievable, just not simple!
wrt object identification and tracking, there are a few layers to this task: firstly “detection” (is there something interesting in the image?) “classification” (identifying/recognising the interesting thing in the image), “scoring” (if there is more than one interesting thing, what is the most interesting thing?), “location” (where is the most interesting thing relative to me and the world?), and then feeding the location of the image to the flight controller in a way it can be consumed. Software and hardware exists that can do all of those things, but doing it in a small, resource efficient way so that it can be done onboard a drone is still hard.
Something like Randy’s balloon finder gets away with simplifying the problem, because the target object is basically uniform and known - so detection and classification can become the same task - and it only looks ahead, basically reducing the tracking problem to 2 dimensions rather than 3.
Distance sensors can work in any direction (ardupilot supports up, down and 8 radial directions, for instance), but unless you get into stereo vision or some very funky algorithms, they are generally active sensors that need to bounce off a target to get a range. If you are going to start looking at some of the more complex options, start looking into ROS as well rather than trying to do everything from scratch.
Regarding building a drone - if you need to hover and fly backwards, go with a multicopter to start with. Helicopters and VTOL fixed wings are cool, but the complexity will distract you from the core task. Simplest is a quadcopter, but depending on the amount of gear you may need to look at bigger (hex/octo): 3 things to keep in mind.
- Real estate. You will need more space to put things than you first realise, so a design with plenty of volume and payload area is important. These might be worth looking into: Dev Frame Development Part 2 and https://hobbyking.com/en_us/s500-glass-fiber-quadcopter-frame-480mm-integrated-pcb-version.html?___store=en_us
- Repairability/availability. It is better to have and not need, than need and not have. You will almost certainly crash a few times. If you build a quad, buy 6 motors and escs at the start, spare landing gear and lots of props. Nothing worse than waiting on parts for a couple of weeks due to an incident when you really need to be testing and collecting data. And buy twice as many batteries as you think you need.
- It’s a tool. Research platforms are different from consumer products. It doesn’t have to be the most efficient, expensive or prettiest drone out there. It just needs to carry your gear long enough to perform the task you are working on. Don’t get caught up with aesthetics.
Just my thoughts, hope that helps!