Notifications
Clear all
Topic starter
2026-07-30 11:06 pm
@inq wrote:
... do I recall that you (@robotbuilder) did a lot with vision recognition. At some point I'd like to try that.
Indeed I have and could talk about visual processing until the cow comes home...
@inq wrote:
... I was brain-storming that if I used two of the ESP32-S3-CAM boards with the higher resolution OV3660 cameras, I could do much of the object recognition on the independent S3's and pass those data streams (of objects, not pixels) to the RasPi, it would only have to process the binocular vision portion to get spatial awareness... and then continue to do the higher order things with that information.
I assume here you mean using AI object recognition on two images one from each of two cameras and using the disparity in the position of the box in each image of the same object?
Probably obtaining a more precise position would be a TOF camera which returns the distance as well as the color of each pixel much like that little sensor you were using on your Inqster bot. Some iphones have that TOF ability built in. (not mine unfortunately).
2026-07-31 9:42 am
would be a TOF camera
Well now... I've missed that development during my absence. They have hobbyist level sensors that do both???
Topic starter
2026-07-31 11:25 am
Not sure of how many hobby level options of any are available.
Obviously if you have a depth map of an image computed from two color images you still have the color images to overlay that depth map image. TOF data is an accurate way to get depth rather than computing it from two images.
Looking online ...
https://blog.arducam.com/time-of-flight-camera-raspberry-pi/
"Standard Time-of-Flight (ToF) sensors only measure the intensity of the reflected infrared light, which creates a greyscale (monochrome) IR image alongside the depth map. However, manufacturers frequently build RGB-D ToF cameras, which combine a standard color sensor and a ToF sensor into a single housing to deliver both color and depth."
2026-07-31 12:54 pm
TOF data is an accurate way to get depth rather than computing it from two images.
Back in those earlier tests I did, you might recall that the infrared range it uses easily got confused with sunlight... even through windows in the auditorium. Never really tried it outside. Was kind of disappointed in the results.
2026-07-31 5:06 pm
I’m afraid I don’t remember what your stated goal was when you were evaluating the ToF sensor, but I’ve been very impressed by the STMicroelectronics Time of Flight sensors. I’ve been able to program a robot to identify table/chair legs via small ROIs & to follow a flashlight based on ambient light readings.
You can program the sensor to drop down into short distance mode when it detects some ambient light threshold and if need be you can also increase the timing budget… They’re great. This is from the VL53L1X data sheet. Were you expecting more?
Tom
Distance Mode
The VL53L1X has three distance modes (DM): short, medium, and long.
Long distance mode allows the longest possible ranging distance of 4 m to be reached. However, this maximum ranging distance is impacted by ambient light.
Short distance mode is more immune to ambient light, but its maximum ranging distance is typically limited to 1.3 m.
|
Distance Mode |
Max. distance in the dark (cm) |
Max. distance under strong ambient light (cm) |
|
Short |
136 |
135 |
|
Medium |
290 |
76 |
|
Long |
360 |
73 |
To err is human.
To really foul up, use a computer.