Hi, I am working on a project to detect spools (reels) as shown below in an example image using the OAK-D 3d camera. I have trained a custom model and it is able to detect the spools. The spool has a very unique design with to circles (flanges) on the sides and a drum/barrel in the center which eventually grows in size as product in wrapped around it to overall look like a cylinder. [upl-image-preview url=https://discuss.luxonis.com/assets/files/2024-05-15/1715790796-248899-image.png] [upl-image-preview url=https://discuss.luxonis.com/assets/files/2024-05-15/1715790877-24838-image.png] [upl-image-preview url=https://discuss.luxonis.com/assets/files/2024-05-15/1715789826-728866-2024-01-19-00-00-00677155-3.jpg] However, in case of multiple spools being detected in a frame, I wanna chose the one which is closest to the camera as that will be the spool of interest to us. Now, given that the camera has 3D capabilities with depth and disparity frames available, I wanted to check with you if you would have any suggestions for doing this using the available 3D data from OAK-D cameras. Thanks, Yishu

Hi @"yishu_corpex"#2472 What kind of model are you using for detection? In case it is YOLO or MBNET, we have integrated spatial inference using [MobileNetSpatialDetectionNetwork](https://docs.luxonis.com/projects/api/en/latest/components/nodes/mobilenet_spatial_detection_network/) and [YoloSpatialDetectionNetwork](https://docs.luxonis.com/projects/api/en/latest/components/nodes/yolo_spatial_detection_network/). Thoughts?

Filtering instances of an object of interest using 3D data

yishu_corpex

Hi,

I am working on a project to detect spools (reels) as shown below in an example image using the OAK-D 3d camera. I have trained a custom model and it is able to detect the spools. The spool has a very unique design with to circles (flanges) on the sides and a drum/barrel in the center which eventually grows in size as product in wrapped around it to overall look like a cylinder.

However, in case of multiple spools being detected in a frame, I wanna chose the one which is closest to the camera as that will be the spool of interest to us. Now, given that the camera has 3D capabilities with depth and disparity frames available, I wanted to check with you if you would have any suggestions for doing this using the available 3D data from OAK-D cameras.

Thanks,
Yishu

jakaskerl

Hi @yishu_corpex
What kind of model are you using for detection? In case it is YOLO or MBNET, we have integrated spatial inference using MobileNetSpatialDetectionNetwork and YoloSpatialDetectionNetwork.

Thoughts?

yishu_corpex

Hi @jakaskerl

Yes, I'm using a YOLO model for object detection. Thanks for sharing the above integrated spatial inference examples. It should provide me with the depth Z for each object that I can use to filter out the objects. I will try it out for my application and will get back to you.

Thanks,
Yishu