Can very small objects be detected by the ZED Cameras?

Last updated: October 23, 2026

The ZED SDK provides the left and right rectified images of the scene and computes the 3D position of each pixel. The smallest object that can be detected depends on its size in the image (number of pixels), which in turn depends on the distance, the focal length of the lenses, and the image resolution. When an object is too far or too small, the depth estimation is not reliable and its pixels can be filtered out by the confidence thresholds.

How to improve the detection of small objects

  • Choose the right camera: for close-range applications, such as pick-and-place, manipulation, and inspection, the ZED X Nano (depth range from 3 cm) and the ZED X Mini (from 10 cm with wide lenses) provide the best accuracy at short distance. For small objects at longer distances, a narrow lens (4.6 mm) increases the number of pixels per object. See For the depth sensing, what is the range for the ZED cameras?

  • Use the highest useful resolution: a higher resolution increases the number of pixels on the object.

  • Use the NEURAL depth modes: they preserve object edges and fine details better; see Depth Modes.

Detecting specific objects

  • Custom object detection: train or use your own detector (for example, a YOLO model) to detect your objects in the 2D image; the ZED SDK then computes their 3D position, 3D bounding box, and tracking. See Custom Object Detection and the YOLO integration.

  • Fiducial markers: stick ArUco or AprilTag markers on the objects you want to track, detect them in the ZED images with OpenCV, and use the ZED depth or point cloud to retrieve their 3D position. With ROS 2, see the AprilTag Detection example with Isaac ROS.

  • Classic computer vision: color segmentation or shape detection with OpenCV, combined with the ZED depth data, can be effective in controlled environments.