Have high hopes for this small camera device, but ...

1. Uses a SparkFun Qwiic connector for the i2c connection. That's fine but SparkFun doesn't seem to make a Qwiic connector to the (much more common) Dupont 2.54mm Pitch Connector cable or adapter.

2. I would be willing to pay significantly more if Useful Sensors added more smarts to the module to make facial recognition more reliable.

3. The camera mounts to the circuit board via a very small connector. Unfortunately I found that the camera separates from the circuit board too easily. I will attempt to put it back but don't have much expectation that I will be successful.

Having said this, I would still say that it is a good buy for $10. Just hoping that Useful Sensors sticks around as a company and version 2 takes into account some or all of the points above.

I bought one a few weeks ago, but it seemed a bit flaky (hit-and-miss on the face detection), so I bought two more, under the assumption that the camera module might be faulty (dead pixels, out of focus, etc). The other two arrived today, and both failed to work at all. Swapping the camera module from the first one to the other two (the camera module is only held in place by its connector) caused them to work, oddly enough.

So, by the looks of it, I've got three working sensor boards but only one functional camera. If I could figure out where to get the cameras (appear to be Arducam PY122), I could possibly repair the two non-functional sensors, but I haven't been able to track down that particular product yet.

Now, granted, Either FedEx or USPS might be to blame here. The shipment of the two latter sensors arrived with a sticker on the box indicating that the package had to be re-barcoded due to damage in transit. It's entirely possible that the camera modules were also damaged during shipment.

Still a fairly disappointing experience, overall.

This novel sensor is a lot of fun to play around with. It performed as expected when using CircuitPython on a Pico (or Pico W) to detect a face and track its movement to change the position of a servo or a visual display (see example projects at: https://www.hackster.io/useful-sensors/products/person-sensor). I was also able to program it to distinguish between different faces (including large pictures of faces), although the accuracy/reliability would likely not be adequate for a security system. The main limitations I saw were that it performed less reliably in low light and that its range is only a couple of meters. The orientation of the sensor relative to the person’s face also makes a big difference, especially when trying to distinguish between different faces. It worked best when the face was well illuminated, within 1 meter, and looking directly at the sensor. When those conditions were met, it worked quite well.

The Jury is still out, sadly, despite the initial enthusiasm.

The simplest Arduino sketch reports (via Serial Monitor) no I2C data!???

The Person Sensor status LED duly lights up when an object is presented in front (including presumably faces).

A vanilla I2C sketch reports that an I2C device is present at the designated address.

What can a newbie do to avoid self-inflicted operator errors?

I guess try it on a Pico next and then of course on the grand daddy -RPi if I can buy without donating to scalpers.

Anyway, good introduction to Qwiic which will serve me well since I like small form factor JSTs and not the Gr___ size.

I wish they didn’t have a proprietary cable for the device. I just peeled the connector housing apart and soldered wire leads directly to the board. The sensor works well under normal lighting conditions. Once there is a bright backdrop it may flare out the camera picture and not get a good detection and/or tracking (understandably for a pinhole camera). It was easy to connect to an arduino, find the library and heading to compile to my sketch.

Bought this to try out for control of an animatronic eye mechanism. The person sensor worked flawlessly after following instructions from YouTube channel "Workshop Nation". Incredible price for what used to be rather tricky tech - but is now plug and play. Here is a video of my test on first power up. https://youtu.be/ernsnKQk_5Q

This is so cool. I've only just started using this little wonder, but now I want to incorporate it everywhere, especially in my smart home projects!

I bought eight of these hoping to use them in a robotics project for a class. Unfortunately, after testing all eight of them, I can say they're not reliable in any way.

I tested them under a variety of lighting conditions. They are more apt to trigger under bright lighting conditions, but false positives become more likely as well. The green light will sometimes indicate when I point the camera at my (or another person's) face. Often it will indicate only once after a delay, and then stop indicating immediately. But it will consistently indicate when switched to a brighter surface (that isn't a face). I was able to get it to consistently indicate on synthetic white fur with barely any pauses.

Hooking it up to a microcontroller, you're able to peek the values, including a box_confidence, (1-100). This would seemingly allow for finer-grained control in order to eliminate the copious false positives, but the number is meaningless. White fur, for example, produces mostly 99 confidence, while a human face ranges between the low 60's and the high 90's from moment to moment.

I'm pretty bummed about this purchase, since I was hoping to use it as a "black box" for a kids' robotics project. It feels like I purchased eight divining rods though. They're nowhere near consistent enough to use in a classroom environment.

I'm building some mechatoric eyes and had absolutely no trouble integrating this sensor into my project. It was almost plug and play. The examples were spot-on and straight-forward.

The person sensor work very well, I am currently using it with and Arduino Mega 2560 for some simple facial detection and it does a great job. Perfect little sensor for a project in detecting a person. While it should not be confused with facial recognition, it does do a good job of detect the presence of a person. Tracking from left to right and up and down is simple. Only thing it needs is to have the cable included with it.

The sensor is fairly easy to use -- except that the Arduino I2C library, which I believe is used in the onboard ESP32, is limited to 32 bytes. The example program shows how to get the other 7 bytes (total packet length is 39) with a second fetch, but this is simply returning a copy of the first 7 bytes of the packet -- it is not the end of the packet (which is the last 5 bytes of face #3, and the 2-byte CRC). The demo program doesn't validate the packet CRC. Had the author done that they would have discovered this problem before releasing the current code.

I am using the Propeller in my project, but ended up running an Arduino demo so that I could connect a logic analyzer to see what's happening. All of the reads from the Arduino demo have the problem with duplicating the first 7 bytes to the last 7 bytes. I believe this could be a firmware error.

It might be wise of the vendor to reduce the maximum faces detected to three which would keep the total packet under 32 bytes and eliminate the buffering problem we're facing now. Another option would be to add the ability to set the read pointer inside the device. ATM, a read is from the front of the packet. You don't get the option to set the address. If this were added then it would be cleaner to read data from the sensor.

Update: Today I ran the Pico demo. It reads an extra byte (40 instead of 39 as in the Arduino code) and it does it all in one go. That said, the last 8 bytes seem to be a copy of the first 8 bytes. This causes me to believe that the Arduino library with the 32-byte limit is being used in the device. It would be nice if a future update treated the device like an addressable memory device where you could read data in chucks of the user's choosing.

I used one of these to create a "Look at Me Light Switch." (https://blog.arduino.cc/2023/03/08/turn-your-lights-on-and-off-by-staring-at-this-little-robotic-switch/) I've just scratched the surface of the capabilities of this device. For face detection at low cost and in simple form, then this is probably the best thing available.

Works reliably event in low light. The inference struct is well thought out. Expect to code in some simple logic to filter out low confidence false positives. Runs 1 inference every 200ms so roughly 5 FPS.

The sensor works fine and the detection rate is good for adults but didn't do too good for children. My only problem was the sensor came with the camera plugged in the wrong orientation and not fully seated in