RealSense has added a close-range mode called Improved Close-Range Depth, or Min-Z, to its Perception Studio software. The company says it extends usable stereo depth to approximately 2 cm on the D401 and D405 cameras and approximately 12 cm on supported D43x and D45x models, with no camera replacement required.
Stereo cameras lose depth up close because the disparity an object needs can exceed what the native configuration supports, and parts of the scene show up in only one imager. Matching also gets harder around edges, reflective materials and low-texture surfaces. Min-Z outputs an enhanced depth frame that downstream applications consume like standard depth data through the RealSense SDK pipeline. RealSense describes the goal as recovering valid depth where native output has gaps, not simply adding pixels near the lens.
Where the processing runs depends on the camera. On D400-series models, Min-Z runs on the host, an NVIDIA Jetson platform, so it shares compute with navigation, planning and inference. On D500-series cameras, it runs on the camera's Gen 5 system-on-chip, and the host receives the enhanced depth directly. RealSense says that lowers the load on host CPU and GPU resources.
For metrology and inspection engineers, the relevant detail is RealSense's own caution. The company lists close-up measurement, dimensional analysis and defect inspection among Min-Z use cases, but says applications with strict metrology requirements should validate accuracy against their specific tolerances. It also treats the 2 cm and 12 cm figures as practical targets, with results shifting by camera model, scene geometry, lighting, material, texture and exposure settings.
Materials matter. RealSense says reflective, transparent or highly absorptive surfaces may produce unstable results. The company adds that Min-Z offers limited value when all relevant objects stay beyond the native minimum range, when a D400 deployment has no spare Jetson compute, or when a purpose-built short-range sensor already meets the requirement.
RealSense points to bin picking as a main use case. Objects often enter the native blind zone as a gripper approaches, forcing developers to fall back on prior estimates, RGB-only processing or extra sensors.
