3 ms·
Can your system make out where the walls and other objects are in the room in real time, and thus make sure virtual objects do not overlap? One striking demo o
by digi_owl 10y ago
Can your system make out where the walls and other objects are in the room in real time, and thus make sure virtual objects do not overlap?
One striking demo of Tango, back when they unveiled their dev platform tablet, was that you could use the tablet to "throw" virtual balls into a random waste basket put on top of a chair. Everything computed in real time based on what the sensors picked up about the world around it.
So if you want to measure up your apartment, all you basically have to do is take a Tango device for a walk around it.
- AndrewKemendo 10y agoYou are basically asking about a few different things - though they are related. Can your system make out where the walls and other objects are in the room in real time, The first thing is plane detection and tracking. The answer is somewhat. We can detect and track some planes, but it's not yet up to par with native depth sensing systems in ideal environments. It performs better in high sunlight environments however. thus make sure virtual objects do not overlap? This is called occlusion. The short answer is, no right now our system doesn't do occlusion. To do that right, you have to build a high probability real time polygon mesh which we aren't building into a live product yet. So if you want to measure up your apartment, all you basically have to do is take a Tango device for a walk around it. There is more than one way to do this. We can do this, but it's not real time - it takes a few minutes. However it's getting faster very quickly and this use case is in the pipeline.
- tlarkworthy 10y agoDedicated sensor hardware blows away laborious fancy math to overcome the limitations of a sensor. Look at the adoption of the Kinect and LIDAR in robotics hardware. No-one is seriously using monocular vision for any kind of serious environmental sensing. Stereo vision + pattern is pretty much the minimum to get satisfactory results.
- nacs 10y ago> No-one is seriously using monocular vision for any kind of serious environmental sensing I believe Tesla uses a single camera for auto-steering.
- AndrewKemendo 10y agoDedicated sensor hardware blows away laborious fancy math to overcome the limitations of a sensor. This doesn't work at consumer scale. No-one is seriously using monocular vision for any kind of serious environmental sensing That's why we're different.