I bought an Xbox 360 Kinect for almost nothing. The plan: point it around my bedroom, capture the whole room in 3D, and end up with a point cloud I can spin around on my laptop. A poor man's LiDAR scanner sitting on my desk.
First honest correction to myself: the 360 Kinect isn't really LiDAR. Real LiDAR fires a laser and times how long the light takes to bounce back. The original Kinect does something different — it projects a grid of infrared dots across the room and watches how that pattern distorts on whatever it hits. The deformation tells it depth. It's called structured light, and it's a clever hack rather than a laser rangefinder. The Kinect One (the Xbox One version) is the one that actually uses time-of-flight, which is much closer to LiDAR. But the 360 sensor is the one I have, and for mapping a room it should still get me a usable depth image.
So here's what I'm setting out to do:
- Get the Kinect talking to my computer over USB. It uses a proprietary connector, so I need the power/USB adapter, then a driver or library that can read the depth stream.
- Pull the depth frames and turn them into a point cloud — every pixel becomes an (x, y, z) point in space.
- Walk the sensor around the room and stitch the captures together into one model of the whole bedroom.
The stitching is the part I expect to be hard. A single capture only sees one wall at a time, from one angle, with a range of maybe a meter to a few meters. To get the whole room I have to register many captures into a shared coordinate frame, and small errors compound as the scan grows. This is the same drift problem real 3D scanners fight.
What I want out of it is mostly just to see my room as a cloud of points — measure the actual dimensions, check if that bookshelf will fit on the other wall without dragging it across the floor, maybe export something I can drop into a 3D scene.
But the more interesting question is what else this thing can do, because a cheap depth camera is a surprisingly general tool. A few applications I want to explore:
- 3D scanning small objects. Same idea as the room, smaller scale — spin an object in front of the sensor and reconstruct it for printing or modeling.
- Skeleton and body tracking. This is what the Kinect was built for. It can track joints in real time, which opens up gesture control, posture checking, or analyzing my swim/bike/run form.
- Presence and motion sensing. Depth data ignores lighting, so it's a robust way to tell whether someone's in a room or to trigger something when I walk past — without a normal camera watching me.
- A cheap sensor for robotics or computer vision projects. Obstacle detection, depth-aware vision, that kind of thing. It's a lot of capability for the price of a sandwich.
I don't know yet how far I'll get, or how rough the point cloud will look once drift sets in. But that's the fun part. Let's see what a fifteen-year-old game accessory can still do.