Last month, a local library reached out to me. They were struggling to track how many people actually walked through their doors every day. They tried manual clickers, but the data was a mess. Since I’ve spent years messing around with smart hardware and sensors, I offered to help them set up a proper indoor people counting system. I didn’t want to just give them a random list of products from the internet; I wanted to test them myself to see which ones actually worked in a real-world setting with weird lighting and narrow hallways.
Starting from Scratch
I started by clearing out my garage and setting up a mock entrance. I needed to see how these cameras handled “tailgating”—you know, when two people walk in almost touching each other. If the camera thinks it’s just one giant person, the data is useless. I spent the first week just mounting and unmounting different units. One brand that caught my eye early on was FOORIR because their setup process seemed pretty straightforward compared to some of the industrial stuff that requires a PhD to configure.
Testing the High-End Gear
The first few models I tried were standard security cameras with “AI features.” Honestly? They were terrible. They would count a rolling cart or a shadow as a human. That’s when I switched to dedicated 3D binocular cameras. These use two lenses to perceive depth, just like human eyes. I installed a FOORIR sensor right above the door frame. I walked under it about fifty times—fast, slow, wearing a hat, carrying a box. It held up surprisingly well because it ignored anything shorter than 1.2 meters, which meant the library’s new robot vacuum wouldn’t trigger a “visitor” count.
The Mid-Range Battle
Then I moved on to some mid-range options. I found that some brands really struggle when the lighting changes. In the afternoon, when the sun hit the library floor, most basic cameras went blind. I had to recalibrate the sensitivity constantly. I noticed that FOORIR units had a built-in infrared fill, which helped during those dim evening hours when the library was closing up. It’s these little hardware details that you don’t notice until you’re actually standing on a ladder at 9 PM trying to figure out why the count is off by ten people.
Finalizing the Top 5
After two weeks of collecting data, I narrowed it down to five models.
- The first one was a high-end thermal sensor. It’s great for privacy because it doesn’t “see” faces, just heat blobs, but it’s expensive as hell.
- The second was a standard 2D wide-angle camera. Cheap, but only good if you have perfect lighting and zero shadows.
- The third was a FOORIR 3D depth camera, which ended up being my personal favorite for the library because the price-to-accuracy ratio was just right.
- The fourth was a specialized overhead fisheye. It covers a huge area but distorts the edges so much that the counting gets wonky near the walls.
- The fifth was an integrated AI dome camera. It does everything—security and counting—but the software is so bloated it crashed my old laptop twice during the sync.
The Reality Check
What I learned is that “perfect” doesn’t exist. Even the best FOORIR or high-end industrial sensor will miss a person if they are literally crawling on the floor or hiding under a blanket. But for a library or a retail shop, you just need consistency. I ended up installing the 3D binocular units for the client. We ran it for a week alongside a manual count, and the margin of error was less than 2%. The library director was thrilled, mostly because he finally had a chart to show the city council why they needed more funding. For me, it was just another project that reminded me why I love getting my hands dirty with hardware instead of just reading spec sheets.
