Sharper than the iPhone!

64 megapixels that's a lot of megapixels and you know what's better full-time autofocus on a pi camera well maybe we'll have to see how this thing works it's articam's new hawkeye camera for the raspberry pi and it costs 36 bucks during their kickstarter campaign they sent me this early production unit and i'm going to see how it stacks up the camera's board layout is identical to v1 and v2 pi cameras but the module on the front is pretty hefty so it won't fit everywhere a normal pie camera does that's because inside the giant metal housing is an electromagnetic autofocus system that allows this camera to have full-time autofocus but it also makes it more susceptible to vibration so use on a drone or certain robots might not be as good as a fixed lens camera and because of that motor this camera actually makes a tiny noise and moves in and out when it auto focuses it's not annoying but definitely worth keeping in mind if you want to build a custom enclosure for it you can use up to four of these things together for quad imaging but you have to have a compute module 4 to do that you can get the full 64 megapixel resolution on a pi 4 or compute module 4 but the resolution drops to 16 megapixels on older pies the actual sensor in here can get 60 fps at 1080p or 20 at 4k but the raspberry pi can't handle that much data so you're limited to 30 fps at 1080p i think this sensors a sony imx 686 but arducam won't confirm the exact model if it is that sensor it has a pixel pitch of 0.8 micrometers which is half the hq camera's pixel pitch and still a little smaller than the v2 camera so the big question is how well the optics and sensor work together and if the camera can still produce a great image with that smaller pixel size but still the sensor itself is pretty big the hq camera used to be the champion here with a sensor that's 7.9 millimeters diagonal but now the hawkeye is the biggest 9.2 millimeters but to use the hawkeye at all you have to install arducam's custom driver that driver isn't compatible with the pi cameras so you can't mix and match or switch between a hawkeye and a pie camera without reinstalling hopefully the driver will be upstreamed at some point but for now it requires a manual install and before we get to testing i have to make one thing clear sensor size megapixel count even specs like the lens aperture don't really matter at all in isolation for any camera system the whole system makes the final image there's a reason pros at the olympics aren't running around with a raspberry pi 4 in one of these arducam shooting sports i have this old nikon d700 and it only has a 12 megapixel sensor but the pictures i take with it will blow away the images i can get out of any pi camera even the hawkeye and that's ignoring every other feature that makes this camera legendary like its build quality weather sealing battery life and interface under certain ideal lighting conditions you could probably coax a pretty nice image out of the hawkeye but don't expect to buy a hawkeye and a raspberry pi and build something on the level of even a cheap mirrorless camera the laws of physics really limit what you can do with such a tiny sensor even if it's the latest and greatest from sony but the hawkeye isn't meant to compete with a handheld mirrorless camera it's made to work for computational photography for surveillance or for other hobby projects it's basically a modern high quality phone camera but one you can control with your own software which is pretty powerful if you think about it like take for instance their stereo and quad camera kits you could build an extremely high resolution 360 degree camera or some cool custom vr rigs there are definitely limitations with current gen pi hardware but i'll talk a little bit more about that later when i get to testing with all that out of the way i went ahead and plugged the camera into my pi 4 model b using the included camera cable put pios on a micro sd card and put that in the pi then mounted the little protective case to a little tripod arm on my desk i booted up the pi and installed the custom articam drivers i put all the instructions for that in my blog post which is linked below the first thing i wanted to test was image quality the user guide recommends shooting at a resolution around 16 megapixels and at that resolution the camera looks great but that's leaving some performance on the table this camera goes up to megapixels this is the same image but taken at a resolution of 9248 by 6944 look at the detail in that screw zoomed all the way in there's a tiny bit of chromatic aberration but it's a lot sharper than the 16 megapixel image and it might look like someone added a tiny bit of haze to this picture but that's actually not haze that's the result of diffraction i won't get into the physics of it but basically when you go for such a high resolution on such a tiny sensor the aperture actually gets in the way causing light to bend a little and that means extreme detail gets a little bit fuzzy even though this lens has an f 1.8 aperture it's not big enough to prevent the loss of some detail at 64 megapixels i can't measure it exactly at my house but i guess the hawkeye's lens is pretty good up to maybe 30 or 40 megapixels past that you'll get that slight haze from diffraction it's not all bad though you can usually sweeten up an image in post-processing and the diffraction won't even be noticeable let's take a look at the same picture taken with my iphone 13 pro its camera only has a 12 megapixel sensor but discounting the resolution loss details are still pretty well defined but wow it is nice to have that additional resolution i kind of wish my iphone had more resolution now but the hawkeye isn't really competing against an iphone it uses the pi camera form factor so let's see how it stacks up against the v2 camera module here's a picture i took with the hawkeye from where i normally record my videos the details are great though it looks like the articam's autofocus picked up on the background in the shot and not my face as i mentioned in my last video on arducam's first autofocus camera the driver that controls autofocus has an algorithm that you can't really modify so it's kind of a black box it's not like my mirrorless camera where i can choose a focus point or use face detect something like that could be possible with the right software but it's really not perfect out of the box but here's the same shot on a pi camera v2 the color and exposure are practically the same but there are a few interesting things first for some reason there was a bit of yellow banding going on on the hawkeye's pictures it's easiest to spot in the background where the vent cover is that banding wasn't visible at all in the pi camera and it might have something to do with my lighting but it's just something i noticed more on the hawkeye second the blacks here especially are way less noisy on the hawkeye than the pie camera the sensor and lens work together to make a much clearer image i mean look at the name badges in the background or at the top of my guitar we can actually see that there's a missing string if you look at the hawkeye's picture moving on to video both the pi camera and hawkeye are limited by the pi's bandwidth i took both of these side-by-side video clips using a similar setup and they look a lot like each other but one thing to note is the cropping on the hawkeye is much less severe so you can get more in frame when recording hd video than you can with a v2 camera but other than that video works similar to other pi cameras and it tops out at 1080p at 30 frames per second i tried 4k and some other higher resolutions but it kept giving me a failed to output streaming warning the camera module on the hawkeye supports higher frame rates and resolutions but no current pi can take advantage i also use some of arducam's custom driver features to test things like continuous autofocus and digital zoom but both of those features need a bit more time to come out of beta continuous autofocus worked but i have no clue what gets it to trigger a refocus you can use a key press option to manually focus but i hope the continuous autofocus algorithm can get a little better and digital zoom also worked but the interface in the hawkeye driver is a little tricky use w and s to zoom in and out the i j k and l keys to move up left down and right and m goes straight to 10 times zoom this feature is sometimes called eptz because you could have a camera set up over a wide area and digitally zoom in and track a subject but it's never going to be as smooth or sharp as true ptz it's still a neat trick though and i can imagine a few cases where some good custom software could put it to use so overall what do i think about this camera well the kickstarter price of 36 bucks is a steal for a camera module this good the software is still half baked and many of the same gripes i had about the 16 megapixel camera apply here and you can go check out my video on that camera for a comparison with a pi hq camera too the final retail price though looks like it'll be around 50 bucks just out of my no brainer range for some projects it's perfect for others you might still consider the hq camera in a good lens instead unless you really need that extra resolution i think a lot of people have a dream of building a pocketable point-and-shoot camera with this i mean you can but the software really holds this thing back autofocus isn't predictable exposure and color can be way off and especially for video the pie just doesn't have the power to make a compelling alternative to a cheap point and shoot but this camera is great for use cases where the pi camera already shines like robot vision surveillance and things like that it really depends on if you need that extra resolution and autofocus until next time i'm jeff gearling

Channel: Jeff Geerling

Views: 547,943

Keywords: raspberry pi, arducam, hawkeye, hawk-eye, camera, pi camera, picam, motioneye, octoprint, octopi, 3d printing, monitoring, surveillance, autofocus, megapixels, photography, photo, video, recording, ai, ml, machine learning, artificial intelligence, robotics, review, how, setup, install, guide, build, custom, 64, 16, pi 4, model b, compute module, quad, dual, rig, csi, embedded, hq, module, v2, performance, low light, detail, sensor, sony, imx686, quality, banding, shutter, control, python, software, vision

Id: 0mYRHrLYmLU

Length: 9min 15sec (555 seconds)

Published: Wed May 11 2022