Good luck detecting a mosquito optically from a distance of several meters using a cheap camera and Raspberry Pi. Oh and you want to do from a moving drone. That will certainly make it work!
Just look at the images in the article - the guy's best result was detecting a black speck appearing on a nearby white wall with some 60-70% reliability (based on his own numbers). So you would be missing a lot of mosquitoes - but will be happy firing the laser at random shadows and what not. And that was in a completely stationary setup and controlled lab conditions, i.e. not at all something resembling a typical poorly lit room!
This article is BS. Preprints are not peer reviewed (i.e. nobody has checked anything in it - so could even be a complete hoax), it is a pretty typical gadgetry style paper (we do it because we can, not because it makes sense) you do at when you need to fill up your resume with research papers (e.g. for keeping/obtaining a job reasons).
The "save the world" (mosquito control, diseases, etc.) justification is also par for the course for this type of crappy paper. Anyone who seriously thinks that one could control mosquito problem by shooting them one by one by a laser is delusional.
But neural networks and "AI" are being used, so it has to be cutting edge groundbreaking stuff, right?
BTW, this nonsense idea has been floated as a publicity stunt a few years ago (including a slow motion video of a laser burning off wing of a mosquito in flight) and it seems that some Russian PhD student from a fairly obscure uni either didn't do their research or has reinvented the wheel (or just plain copied the thing without attribution). The list of irrelevant or only very tangentially relevant (it is about mosquitoes, so in scope, right?) references is a dead giveaway there (paper on mosquitoes spreading zika? seriously?).
Oh and that was supposed to be a handheld device to boot. With the same "save the world from malaria" spiel too. I wonder what are the owners of the company that was pushing this concept to investors back then trying to sell today ...
There are actually multiple videos on Youtube showing products from different companies that were attempting to push this as some sort of viable concept.
>> This article is BS. Preprints are not peer reviewed (i.e. nobody has checked anything in it - so could even be a complete hoax), it is a pretty typical gadgetry style paper (we do it because we can, not because it makes sense) you do at when you need to fill up your resume with research papers (e.g. for keeping/obtaining a job reasons).
That describes about 80% of the field of machine learning today: that's how most new work in machine learning is presented to the community, through preprints on arxiv that never get published, therefore peer-reviewed; and most of it is of the "we did it because we can, not because it makes sense" type. The same goes for much of AI research in the past. Here's John McCarthy:
1. Much work in AI has the ``look ma, no hands'' disease. Someone programs a computer to do something no computer has done before and writes a paper pointing out that the computer did it. The paper is not directed to the identification and study of intellectual mechanisms and often contains no coherent account of how the program works at all.
I'm prepared to bet that the exact same work could be published by a respectable research team (you know, with good English) and it would get many adulatory comments in social media (though I hope you would retain your skepticism).
even if you had a RED Komodo feeding uncompressed 4K DCI 60fps video to a pci-express bus capture card, the sensor resolution and tiny size of mosquitoes means that unless the lighting conditions are just right, and the mosquito is somehow highlighted against a background, it's going to be very hard to pick them out at distances of 2 or more meters.
and that's before you get into the software problem of processing the fire hose of data that is 4096x2160 at 60fps raw. and the hardware cost of a very serious workstation class PC capable of taking the capture at 1:1 realtime.
possibly a lidar based sensor or something might be more suitable to locating the x/y/z position of mosquitoes in a few meter area.
Mosquitos, as we all know, have a highly distinctive auditory signature.
A phased microphone array is the only sensible approach to localized mosquito detection. It would probably work reasonably well.
The problem is that the entire field is patent encumbered, because Myhrvold's company Intellectual Ventures has done some research on it, and no on in their right mind would go up against those guys.
A LIDAR is not the right solution. A LIDAR misses most of the space a few meters away, and even if by chance the beam passes right over a mosquito, it will be filtered out by the built-in denoising algorithms.
Audio would surely be the way to go, IMO, although I would avoid Bluetooth and generally any non essential digital transport both because of latency and possible compression artifacts. One could trick the mosquitoes into flying through a small restricted area in which wires wouldn't pose a problem by making it the only possible way to some bait for example, then setting mikes and traps in there.
Or you could just spray RAID or something similar and kill all the mosquitoes there for pennies. No laser, no bluetooth, no microphones.
And if you want to actually solve the mosquito issue, don't have standing water around. Or make sure fish and frogs can live in it. They love eating mosquito larvae.
Not this boondoggle. Remember, engineering is also about realizing that the fact that you are able to do something doesn't always mean you should.
To be fair though, part of that is because they hunt at night.
There are a few places here in Sydney where you see a remarkable cloud of bats feeding in bright light - the lights over the harbour bridge and several of the tax buildings bordering the Botanic Gardens regularly have flocks of bats whirling around presumably attracted to all the insects flying in the bright lights...
The paper claims "1 - 5mm" for a mosquito, which seems plausible. Lets pick 5mm out of pure optimism. The new(ism) HQ Raspberry Pi camera modules are 12megapixel ~4000x3000 pixel sensors, if you assume you could recognise a mosquito based on it's flight/movement pattern rather than needing to have an image of the mosquito, maybe having a 2x2 pixel resolution of it might be enough, so you could use an appropriate lens and cover a 5mm x 4000pixel / 2 = 10,000mm or 10m wide by 7.5m tall region. That seems totally do-able. Even up to 4x4 pixel per mosquito at 5m x 3.75m would happily cover the wall of my bedroom with the windows. You _might_ even be able to do that with the cheaper 8 megapixel V2 camera module instead of the HQ one.
The background subtraction is fairly easy to do in OpenCV. Assuming there's not too much moving in the background (I might need to tie back my curtains?) I wonder if the path a mosquito traces out would be distinct enough to detect?
Research (read "the first google hit) suggests mosquitos fly at 1.6-2.4kmh - call it 2 - which is just over half a meter per second, so about 20mm per frame at 25fps, or 8mm per frame at 60fps. If we choose to use the 4px = ~5mm numbers from above, that means if you detect a thing that might be a mosquito, you don't need to search very far around it in the subsequent frame to track it - if it's a mosquito, it's likely to be still there and within 6 or 7 px of where it was in the previous frame. (I've never tried doing this on something so small, but that technique works really well tracking people across multiple frames in a video. See: https://data-flair.training/blogs/python-project-real-time-h... )
I don't think that at least x,y detection of a mosquito in a 12 or 8 megapixel camera is impossible. The 8MP raspi camera has a 63deg horizontal field of view, so that 5m wide assumed view from earlier would be at 5m distance give or take. If you could get your shoot-down laser close enough to your detection camera, you might not need to care about the z axis. The camera module is about 20mm square - at 5m range shooting the laser across the edge of the camera module pcb might be "close enough"? Might not work at short range, though once you're close you could scan the bean around a bit, it'll be super obvious in the camera view when you've got a hit (assuming the background doesn't;t reflect the laser brighter that a lit up mosquito? Great, now I have an excuse to paint my walls with VantaBlack...)
Another idea... Assume a similar setup - a camera aimed at a 5x3.75m cross section of space, with a line laser shining across it (one of those lasers shone through a cylinder that makes a line, like some power tools use to project guidelines). I'd guess it'd be possible (maybe even easy) to distinguish between dust mites and mosquitos lighting up as they cross the line laser? Perhaps have three co-planar line lasers 10mm or 20mm apart to make the timing of aiming your shoot-down laser easier, use the first flash as a detection signal and swivel your galvos to that region, and then use the 2nd/3rd flashes close by to quickly fine tune the aim without needing to swing too far?
(Now I just need to work out which if any of my half finished projects I should stop working on and half finish this one... ;-) )
Great post. I think line laser though is just a point source that spins, creating an illusion of a plane. So it only looks like a line but is actually a moving point. So if the mosquito travels its own length+the line thickness in a time that is less than the time between laser passes, it won't be touched by the laser.
Maybe some line lasers are made like that (now that you mention it I do recall seeing tripod mounted lasers which are probably spinning POV things), but I’ve got a bunch of line and cross laser modules which 100% do not have moving parts. I’m not sure why I think this, but I understand they’re made by just putting a glass cylinder in front of the laser diode module, working as a kind of 2D lens... Pretty sure that’s what you get in jigsaws and circular saws as well?
Shrug, add a nice zoom lens to make it's effective range several meters instead. Of course it would be far sighted but maybe you know most mosquitos will be several meters away anyway. Or have 3 cameras, one for the cm, one for the m and one able to zoom many meters to look at known problem spots like puddles or a bird bath.
Or use a better method than a cheap camera. Like a radar and microphone like the other more famous project does. Because mosquitos have a distinct frequency of flapping their wings. Meaning you can then actually target the biting type. Because .. 95% of the mosquitos around actually do not sting humans(number from the back of my head from a friends paper some years ago). And killing them all has a quite negative impact on biological diversity.
Be sure to include some laser-resistant capabilities in the engineered mosquitoes, so that consumers will be forced to buy upgrades to more powerful lasers.
Yes, at 300mm range, it's just a toy, and mounting a 1W laser with a targeting system on a drone is a terrible idea. I can see ways to make this work, though.
You need a good way to tell that you're on target. The way to do that is to use the vision system only as a search radar, to find that there's something to shoot at and approximately where it is. Then point the laser near the target, at low power, and start scanning around the target. Modulate the outgoing beam, so you can see when it's illuminating something. Get range from time of flight. When you find something worth shooting, go to high power and take it out.
This is roughly how radar-controlled anti-aircraft gun systems work.
An ordinary UV lamp with bug zapper will probably be more effective.
I saw those videos. The problem is the maintenance, adding yeast and cleaning the trap. Imagine scaling to multiple traps in a house or the garden.
On the other side, burning methane generates CO2: CH4 + 2 O2 → CO2 + 2 H2O. Methane is distributed nearly everywhere in my country for cooking and heating. That at least solves the problem of refilling.
I guess we should burn a very small amount of methane to lure mosquitoes into those traps. But how about climate change etc.?
> You need a good way to tell that you're on target. The way to do that is to use the vision system only as a search radar, to find that there's something to shoot at and approximately where it is. Then point the laser near the target, at low power, and start scanning around the target. Modulate the outgoing beam, so you can see when it's illuminating something. Get range from time of flight. When you find something worth shooting, go to high power and take it out.
Not only more effective, but i imagine it would be easy to misdetect someone's iris as a fly D:
> Good luck detecting a mosquito optically from a distance of several meters using a cheap camera and Raspberry Pi.
Same thought as well. They must not have done the math on the optics... You'd have a much better chance using an acoustic array (mosquitoes put out a very distinctive sonic frequency), or even better an array of radar modules. There is simply too much noise and data in the visible spectrum to catch something that small without expensive optics and an expensive processor to process all of those pixels.
This. Acoustic array sounds like a great solution for finding that pesky mosquitto that is hiding somewhere in the room. Anyone know how difficult it would be to make this?
Well, you can get a radial acoustic 7 array of MEMS microphones for ~$50, doing stuff with the output is the hard part. That being said, I've used one to localisation the direction that lightning is coming from. Have no idea how good the accuracy would be, but there's likely large room for optimization in that setup; it likely just comes down to the accuracy of the microphones.
I personally would go with the radar route; it's slightly more expensive that the acoustic route, but accuracy would be fantastic, even at a distance.
With a proper lens I think it just might work. You need to get huge focal length to make those pixels work. Then you need illumination because with those focal length it will be really dim. Then you need to scan the area because your FOV is tiny (but at least you'll get something like 90FPS @VGA res which is enough for AI) so you should be able to see those mosquitoes shining in the air like dust particles in sunlight coming from the window.
But then using a laser just insane - anything that has enough energy to down a mosquito will have enough energy to blind someone unfortunate to see the laser or its reflection. I'd go with power LED and some focusing with very narrow focal depth. maybe a lens moved like in optical drives, maybe a MEMS reflector like are used in projectors (but instead of spreading all pixels to cast a projection focusing it in a single point?
I've integrated a few laser engravers and laser-cutting CNCs, so let me clear up a couple misconceptions: Lasers typically have a narrow focal length, just like your LED proposal - otherwise they'd damage the optics/mirrors! They're designed to pass a coherent beam of about 10mm diameter through a series of lenses and mirrors. The beam of an engraver might carry 10W of power in that diameter, which is pretty bright, but decent lenses or mirrors with high transmission and reflection won't be damaged. It's safe for both skin and optics until, say, 100mm after it goes through the primary lens. Hopefully at the surface of the thing you're cutting or engraving, it's a spot size of 0.1mm! With 10W of power in that tiny area, even exotic antireflective glass that can transmit 99% of the light will be atomized.
If the thing you're cutting/engraving isn't present, then 100mm past the focal plane, it's back to 10mm diameter, and 100mm after that, it's 20mm, and just continues dissipating. It's really pretty hard to focus it back into a coherent beam sufficiently narrow to damage eyesight (much less skin).
Yes, the CDRH is really concerned about the possibility of a laser system accidentally lasing a blob of metal into a perfect mirror that will send a coherent beam directly into the eye of an operator standing meters away; I've got laser systems with thousands of hours on them etching reflective foil off of plastic and there's no sign of reflections causing damage to the smoked acrylic that's shrouding the operation.
The typical mechanism used to drive laser engravers is more like a HDD actuator that moves the read/write heads - it's called a galvanometer. Basically a mirror on a stick connected to a magnet. A voice coil pulls the magnet +/- 10 degrees or so; They've got sub-millisecond response times so hitting a moving mosquito would be no problem!
I propose that you suck the mosquitoes through an enclosure with something like a box fan or duct booster. Make sure the entrance and exit has to go through a couple right angles so your laser optics are never visible to someone outside the box. Backlight the inside of the box so that instead of AI to see whether the thing in your camera FOV is a mosquito or a songbird really far away, you just zap any black mosquito-sized blobs that pass through.
I just played a bit with CNC engraving and laser projection, so don't have that extensive knowledge, but AFAIK:
Even with your example of 10W, 10mm beam it has a bit more than 0.12 W/mm^2. Hundred times the power output from the sun. Even if we use a 1W laser, which should be enough for a mosquito, it's still like looking into the sun with a 10x magnifier.
And as you said - you focus the spot in 0.1mm with a small focal length - that's great for cutting or engraving but won't deliver enough energy to a mosquito 5 meters away unless you start with the same beam size, then focus it 5 meters away at the target. Only then it will need another 5 meters for the beam to return to the initial diameter and many more times that to finally be safe.
What I propose is a big reflector and aiming optics directed at it(a bit like a satellite antenna). I didn't do the math, but tens of centimetres or even more than a meter of dimeter. Let's say 50cm==500mm.
Then if you want to focus it to 5mm at 5 meters it will have less than 10mm at +-5cm. You could do the same with a laser but you just don't need to. Instead of 1W laser you can use a 10W powerled, have the same energy at the target but it will be much safer at almost every stage before it.
And what most people think about when they talk about shooting mosquitoes with a laser is, I think, sending a coherent 5mm beam directly - e.g. from a laser pointer. So you don't have to estimate the distance, but it is even more dangerous.
> But then using a laser just insane ... I'd go with power LED and some focusing with very narrow focal depth. maybe a lens moved like in optical drives, maybe a MEMS reflector like are used in projectors (but instead of spreading all pixels to cast a projection focusing it in a single point?
No a laser is coherent. It has almost the same intensity in it's entire path. If you know the approximate distance you can accurately focus a normal beam of Light into that area. Killing the bug but much safer.
This would protect people standing behind the bug, but if the system mistakenly identified an eye as a bug you would still be delivering a burning quantity of energy focussed on the eye.
Not completely BS for me. I have actually learned something:
> The size of a mosquito can vary from 1 mm to 5 mm, this is the main criterion for the method of detection and retrieval of mosquito coordinates. When monitoring the position of the mosquito with ultrasound, it is necessary to use several sensors in different places and processing their information to calculate the location, which theoretically is suitable only for detecting one mosquito, but if there are several mosquitoes, the device will not work correctly. The temperature of the body of a mosquito due to its cold blood is like the temperature of the environment. With a very high resolution of the thermal imager, the mosquito temperature will differ insignificantly against the background - on the order of 0.1 C. The use of sonar has several difficulties when working in open areas where it is necessary to use sonar with a narrow beam and a narrow radiation pattern.
tl;dr, ultrasonic/infra-red/sonar don't work either.
Just look at the images in the article - the guy's best result was detecting a black speck appearing on a nearby white wall with some 60-70% reliability (based on his own numbers). So you would be missing a lot of mosquitoes - but will be happy firing the laser at random shadows and what not. And that was in a completely stationary setup and controlled lab conditions, i.e. not at all something resembling a typical poorly lit room!
This article is BS. Preprints are not peer reviewed (i.e. nobody has checked anything in it - so could even be a complete hoax), it is a pretty typical gadgetry style paper (we do it because we can, not because it makes sense) you do at when you need to fill up your resume with research papers (e.g. for keeping/obtaining a job reasons).
The "save the world" (mosquito control, diseases, etc.) justification is also par for the course for this type of crappy paper. Anyone who seriously thinks that one could control mosquito problem by shooting them one by one by a laser is delusional.
But neural networks and "AI" are being used, so it has to be cutting edge groundbreaking stuff, right?
BTW, this nonsense idea has been floated as a publicity stunt a few years ago (including a slow motion video of a laser burning off wing of a mosquito in flight) and it seems that some Russian PhD student from a fairly obscure uni either didn't do their research or has reinvented the wheel (or just plain copied the thing without attribution). The list of irrelevant or only very tangentially relevant (it is about mosquitoes, so in scope, right?) references is a dead giveaway there (paper on mosquitoes spreading zika? seriously?).
Here, it was even on National Geographic in 2010(!): https://www.youtube.com/watch?v=BKm8FolQ7jw
Oh and that was supposed to be a handheld device to boot. With the same "save the world from malaria" spiel too. I wonder what are the owners of the company that was pushing this concept to investors back then trying to sell today ...
There are actually multiple videos on Youtube showing products from different companies that were attempting to push this as some sort of viable concept.