Best Quadcopter for Machine Vision Applications

So I pose the question to the community; which quadcopter, out of all on the list, is the most stable, most supported in terms of existing documentation and software, and, do you think, would be the most appropriate for a machine vision application?

The 'application' itself would be a camera payload, mounted to the drone's undercarriage, with a variety of sensors and processors embedded inside. The payload would need the ability to control the drone via it's existing transceiver, so a well recorded protocol with an existing library is a must.

No offense here man but if you're asking for a RTF / ARTF / BNF quad to do this I dont think you'll get very far with your project. Your best bet will be to start with an OSH flight controller- multiwii or another, and develop that software into what you need.

There has already been lots of people (or more correctly corporations) that have machine vision controlling fleets of multirotors (look at the lexus commercials for one), but it's going to be WAY HARDER to have the MV (Machine Vision) on board the craft. Probably to hard (and by that I mean it will require to much HW / code for the controller to be light enough to fit on a multirotor and still fly)

What exactly is it you're trying to do? Just have a MV setup fly a quad around a specific area? That wouldnt be to hard at all (if you're into machine vision already) to setup stationary cameras that track the MR as it moves around and even control it, but to put the HW and needed FW on board is going to be an absolutely huge task, go ahead and get your walled out now and throw it at your computer screen, cause that's what it's going to end up being...

In your shoes I'd look for a supported model that already carries a camera or other accessory (e.g. Syma X5C-1 or WlToys models). You'd get an easy way to power your payload and one or two control signals to it. And the quad would be "designed" for the extra weight, but that's just a few ounces for an inexpensive off-the-shelf quad. What are the weight and power requirements for your payload?

Firstly, thank you both for replying, both valid points.

I definitely understand the scepticism Cereal_Killer, because prior to my training 'machine vision', to me, meant a fully-fledged computer running something C++ based like OpenCV or another similar library - very labour intensive, expensive and heavy. Thankfully, I've been fortunate enough to spend a year working on FPGA-based camera processing, which means I can do things like track pre-determined colours over distance, and closed-loop motor control/PID control without software. Essentially, I can take the input from a camera at just under 30 FPS and determine the exact direction and altitude of an object with hardware weighing just over 50 grams, not including whatever I decide to print an enclosure out of.

Hexfet, if we round up to 100 grams and 3.3 volts with a drain of about 200mA is that sounding achievable? The previous drone I picked has a camera mounted to the bottom and a separate control interface to activate it, which was going to come in handy, unfortunately it's a UDI branded model and I'm reliably told by PhracturedBlue and victzh that the protocol for it's TX is still out of reach, creating the need for a much more 'open' and supported brand and model.

It's not that the protocol is out of reach, I'll try to help you with protocol. But U818a will not bear 100 grams of extra load. You need small brushless quadcopter for this, 200-250mm size.

I suppose that renders the protocol irrelevant then? Is there such a drone on the supported list?

Actually, that may have been a bit of an assumption - at full throttle it pulls 178 - 183 grams, at half it pulls 100 grams.

The 200 qx is the closest I can think of, but this reviewer suggests a max payload around 50 grams. Blade also makes the 350 qx . These use the DSM protocol.

kr0sh1 wrote: Actually, that may have been a bit of an assumption - at full throttle it pulls 178 - 183 grams, at half it pulls 100 grams.

How does it feel if you fly it with 100 grams attached? Would it have the performance you need?

Scrub that, this is it with 100 grams attached! If I'm headed toward brushless motor territory then a higher price-bracket might be an idea - it needs to comfortable carry that kind of weight so I might have to start thinking about kits etc?

I think there's something a lot of answers have overlooked. The smaller quads have integrated RX/flight control boards, and so accessing control channels beyond the flight control channels will depend on reverse engineering the hardware connections for them (i.e. - controlling the WLToys camera, for instance). Even worse, there's probably no way at all for your on-board gear to control the quad.

This is actually similar to a project I'm just now staring on to use a microcontroller + a bunch of sensors to change the behavior of the craft, or to the stand-alone flight stabilizers that are available for helicopters and airplanes. They sit between the Rx and the actual controls and modify things.

You want an aircraft with separate Rx and flight control boards, so you can put something in between them to control the aircraft. Doing this also generally means a larger aircraft, as the RTF/BNF aircraft don't start showing this until you get up to at least 350 size aircraft.

A PNF aircraft by definition has that separation, as you should just have to install the Rx and then fly it. For an (A)RTF/BNF aircraft, I'd check the Blade 350QX3 and the DJI Phantom's. Again, the crucial point is that the Rx and the flight control board are discrete units that you could swap out for some other vendors hardware. I haven't had much flight time on the V3 firmware, but the DJI was the more stable of the two, while the 350QX had more performance.

Since the quad flight controller has to have a microconroller on it (you can't fly a quad except "by wire"), using something that's open source firmware, if not hardware, open's the possibility of modifying the firmware so your sensor gear can communicate with it, instead of having to tweak the signals between the Rx & FC. You might find a PNP kit (i.e. - you just need to plug, and possibly bolt, the parts together to be ready to fly) with one of these as well. Since they are usually frame+electronics, mounting more hardware on them should be straight forward. The HK SK450 has gotten good reviews, and the KK2.1 FC is open source. And it will have plenty of lifting power.

If you have no flight experience, you might try buying an inexpensive RTF quad to practice on. The Syma X5C comes highly recommended for this purpose.

mwm, that's an excellent answer. Which is why I've gone all out and bought this kit , and this flight controller - the kit gives me all of the electronics and frame with no fuss, the controller is open source and can be interfaced to a microcontroller using PWM signals, everything else is up to me.

I looked at the KK2.1.5, have you used it before? I only went with the Naze32 because it's ARM-based and therefore a little more familiar.

Take a look at Cleanflight firmware for the Naze. They've recently integrated data logging which might fit well for feeding sensor and controller data to your payload. The Full Naze version also has magnetometer and barometer sensors. Easier than putting sensors on your payload if you need them.

The Naze32 is a great choice - it gets good reviews. If you're going to be modding it, it's probably a better choice than the KK2.1.5, as it's got more power for handling extra tasks. I've also heard good things about the cleanflight firmware hexfet mentions, though it's not quite as universal as for the Naze32, probably because a lot of people are happy with the firmware that ships with. I choice a KK2.1.5 for my one DIY quad, because I wanted to be able to configure it without plugging it into a computer. But that's not for the project I mentioned, but so I can experiment with 3d-printed multirotor frames.

The Quantum Venture, I don't know about. The ESC's are good, and MultiStar motors OK. You'll need to add an Rx as well as the FC. Given where we are, I assume you're going to use a deviationTx Tx.

Note that FPV and aerial photography have radically different requirements. FPV (and flight videos) use a fixed camera pointed forward, and often upward. Many of them started life as CCD security cameras with small boards. Aerial photography wants a stable platform to shoot from, with a gimbal to compensate for aircraft motion and camera pointing. Not sure which you want, but it's not clear you can put a gimbal under that kit on either set of landing skids.

Good tip hexfet, I was wondering which firmware to go for, hopefully CleanFlight is fairly adaptable as I'll be asking it to do some weird stuff.

And mwm - a feature that encouraged me about the Naze was the gimbal servo outputs. The payload is going to be attached to a gimbal where all the collision detection and visual recognition sensors will be housed on the end of. This eliminates rotational issues, which affect the abilities of my MV design, and allows for consistent viewing angle values to be switched between by adjusting the gimbal presets.

I am really happy with it, it's STM-based so it's pretty close to home, but with the raft of useful sensors, like you say, it takes some of the weight off my payload. As I go along I'll post some images of my work as it progresses and they'll be available on my Hackaday.io page.

Forget to mention, to mwm, I'll print an adapter and whack it on Thingiverse

More FYI www.rcgroups.com/forums/showthread.php?t=2154329 You would have all pins broken out.

Very nice! There's some definite inspiration in there, thanks for that SadSack (never thought I'd say that in a sentence).

Have a look at his other threads lots of things done another way...