Hello im going to use 3d Intel sensor to inspection dimensions and tracking to pick by robot. Which sensor are You recommend for this purpose ? I was thinking about lidar l515 but after Your review I'm stuck :(
that really depends on the requirements and environments of your system. These Realsense devices each have different working distances, precisions, and update rates (frame rates) for depth data. If your inspection for dimensions is targeting for a few mm precisions for objects like in this video, any of the currently known realsense devices might not be a good choice. If that's your case, you might want to consider ToF cameras, like Microsoft Azure Kinect, which is pretty much on the same price line with L515 yet gets you noticeably more precise depth data. Stereo vision devices like in this video have tendency to miss sharp edges. However, ToF cameras have tendency to end up having partially inaccurate scans in tightly surrounded environments. If that's your how your environments look like, stereo vision devices like in this video could rather be a slightly better choice as they don't possess strong scan-deformation sort of characteristics. but really, it does also depend on how the inspection is done. If you are thinking of surface-to-surface measurement from pointclouds, for instance, all of the known realsense devices might just work alright in close range like 0.5 m. Object picking, on the other hand, can often be done with low precision depth scans. I've seen some object picking systems utilizing realsense devices (D435) under limited circumstances. This is presumably because object detection is first done in RGB image, effectively segmenting object points. Solid State LiDARs like l515, though I'm still testing this one, has better tendency to preserve geometric premitives of objects in further distances like 2 m or so. Specifically for l515 though, it has narrower field of view compared to other realsense devices, and I'm kind of seeing larger deviations on plane surfaces compared to D455 (post-processed). Tracking is another thing if you mean tracking of moveing objects/camera, though trackings are commonly done from RGB images these days. so yeah, it really depends on your system design and requirements, so it's hard to say to pick one from your question.
According to the timeline 4-35, you didnt't use spatial filtering for one of the cameras (checkbox was turn off). Maybe that can also lead to the different results
good point. I'll check if that makes much difference. I might have used wrong screen shot of the settings which isn't the one that's actually been used by mistake.
Great work comparing these two !
thankyou for this test
great job！
Which depth camera would you recommend if cost is no issue for highest depth accuracy / frame rate?
Hello im going to use 3d Intel sensor to inspection dimensions and tracking to pick by robot. Which sensor are You recommend for this purpose ? I was thinking about lidar l515 but after Your review I'm stuck :(
that really depends on the requirements and environments of your system. These Realsense devices each have different working distances, precisions, and update rates (frame rates) for depth data.
If your inspection for dimensions is targeting for a few mm precisions for objects like in this video, any of the currently known realsense devices might not be a good choice. If that's your case, you might want to consider ToF cameras, like Microsoft Azure Kinect, which is pretty much on the same price line with L515 yet gets you noticeably more precise depth data.
Stereo vision devices like in this video have tendency to miss sharp edges.
However, ToF cameras have tendency to end up having partially inaccurate scans in tightly surrounded environments. If that's your how your environments look like, stereo vision devices like in this video could rather be a slightly better choice as they don't possess strong scan-deformation sort of characteristics.
but really, it does also depend on how the inspection is done. If you are thinking of surface-to-surface measurement from pointclouds, for instance, all of the known realsense devices might just work alright in close range like 0.5 m.
Object picking, on the other hand, can often be done with low precision depth scans. I've seen some object picking systems utilizing realsense devices (D435) under limited circumstances. This is presumably because object detection is first done in RGB image, effectively segmenting object points.
Solid State LiDARs like l515, though I'm still testing this one, has better tendency to preserve geometric premitives of objects in further distances like 2 m or so. Specifically for l515 though, it has narrower field of view compared to other realsense devices, and I'm kind of seeing larger deviations on plane surfaces compared to D455 (post-processed).
Tracking is another thing if you mean tracking of moveing objects/camera, though trackings are commonly done from RGB images these days.
so yeah, it really depends on your system design and requirements, so it's hard to say to pick one from your question.
According to the timeline 4-35, you didnt't use spatial filtering for one of the cameras (checkbox was turn off). Maybe that can also lead to the different results
good point. I'll check if that makes much difference. I might have used wrong screen shot of the settings which isn't the one that's actually been used by mistake.