4 ms·
Typically the LIDARs that are on autonomous vehicles are pretty low "resolution" compared to smaller LIDAR packages like a Hokuyo or depth sensors like a Kinect
by dljsjr 6y ago
Typically the LIDARs that are on autonomous vehicles are pretty low "resolution" compared to smaller LIDAR packages like a Hokuyo or depth sensors like a Kinect. It wouldn't be good for fine detail scanning.
- echelon 6y agoAzure Kinect isn't great at resolution, either. Its depth sensor is an embarrassingly small VGA resolution, which sucks for realtime photogrammetry. Point cloud scenes look like sludge. Are there any "real time" depth sensors with high spatial and sensor resolution? I've been looking at FGPA-based solutions like Nerian Scarlett, and this seems to be closer to what I want [1], but still not quite good enough. [1] https://nerian.com/products/scarlet-3d-depth-camera/ https://nerian.com/products/scarlet-3d-depth-camera/
- dljsjr 6y agoThe reason I put resolution in quotes is that I wasn't referring to the image/buffer resolution but rather the ability to resolve fine details. For example, Hokuyo LIDARs can resolve details w/ ~millimeter resolution at close range (less than 10m). Another way to think about it is if you want to map your point cloud as a set of voxels in an octree, the smallest voxel is about 1m x 1m x 1m. Conversely, large LIDARs for vehicles like Ibeo and Velodyne and such typically map things with a resolution in the range of centimeters or larger. When you're looking for large things like other vehicles, people, obstacles, etc. then you can make those tradeoffs in the sensor design and fall back on things like vision for small feature detection in a more cost effective way.