LingBot Enhanced Depth Filter for Gemini 330 Series

Powered by LingBot-Depth, the LingBot Enhanced Depth Filter is specifically adapted, licensed, and activated for supported Gemini 330 series cameras. It combines the native depth data from Gemini 330 series cameras with the latest LingBot-Depth capabilities to reduce depth noise, fill missing depth regions, and refine object boundaries. The workflow has three steps: the Gemini 330 series camera captures raw depth and RGB information, the Enhanced Depth Filter enhances missing or incomplete depth data, and the enhanced depth output is delivered for AI and robotics. It is delivered through the Orbbec SDK v2 as a native SDK expansion, implemented via a streamlined Enhanced Depth Filter Class wrapper for rapid client integration.

The Enhanced Depth Filter supports the Gemini 330 series, specifically the Gemini 335, Gemini 335L, Gemini 336, Gemini 336L, Gemini 335Le, and Gemini 335Lg. On the compute side it is optimized for the NVIDIA Jetson Orin series and is fully compatible with the Jetson Orin Nano, Orin NX, and AGX Orin compute modules. The software requirement is a hard constraint: it strictly requires JetPack 6.2 and its underlying CUDA and TensorRT compute stack.

Raw depth data commonly suffers from high noise and scattered points, unstable depth edges, missing data and holes, and lower reliability for AI and robot operation. The Enhanced Depth Filter addresses three difficult scene types. For transparent objects it improves transparent-object integrity, produces cleaner object silhouettes, and delivers more reliable object perception. For low-texture surfaces it reduces noise, produces more complete and consistent surfaces, and improves scene stability and reliability. For edges it reduces holes and missing data, produces sharper and more stable edges, and improves object definition and recognizability.

The Enhanced Depth Filter supports 640x480, 1280x800, and 1280x720, with 640x480 recommended as the preferred resolution. Before adopting the Enhanced Depth Filter, D2C alignment between the color and depth images must first be completed. At 640x480, hardware D2C is supported, allowing the alignment to be performed on the camera side and reducing CPU usage on the host. For other resolutions, software D2C alignment can be performed using AlignFilter. Actual performance may vary depending on the selected resolution and alignment method.

Orbbec published test results measured at 640x480 on an NVIDIA Jetson AGX Orin 64GB Developer Kit in maximum power mode with clocks locked at maximum frequency, using SDK V2.9.2 and the model.sm4 model at 240 MB. Under those conditions, single-core CPU utilization was approximately 109%, GPU utilization averaged 86% with a standard deviation of 15%, and latency was 43 ms. Actual figures will vary with the platform, resolution, and alignment method selected.

Each Gemini 330 series device requires its own unique license, and a free trial is available. First, submit the license application form with your contact information and device details. Second, the Orbbec team issues your License Info in the form of a code. Third, download the Quick Start Guide document. Fourth, get the Orbbec License Tool and follow the guide to complete activation.

In logistics and warehousing, across AMR, robotic, and sorting systems, it supports carton and box edge tracking by providing stable depth boundaries for carton detection, tracking, and pose estimation. In industrial automation, across robotic arm and assembly systems, it supports reflective part alignment through improved depth perception for metal, glossy, and reflective components. In commercial applications for embodied and humanoid robots, it supports transparent vessel handling across retail, healthcare, research, and other settings, improving a robot's ability to perceive transparent objects such as glass cups, test tubes, and medicine bottles for more stable and accurate grasping.