Seeed Hardware: reServer J2032 with NVIDIA Jetson Xavier NX 16GB
Application: Depth Video Data Capturing and Management & Environmental Perception
Background
The world of machine learning can be a complex labyrinth, especially when it involves decoding visual data from depth images in a 3D environment. It requires the perfect blend of accuracy, stability, and automation. At Aivero, we’re working at the cutting edge of this field, leveraging machine learning and advanced AI to manage and process massive volumes of user-generated depth videos.
We’re excited to announce our collaboration with Seeed Studio, a leading provider of NVIDIA-based server and edge AI hardware. Together, we aim to revolutionize the way 2D and 3D visual data from depth cameras are captured and handled.
The Challenges in Machine Learning and Computer Vision
Implementing computer vision and machine learning automation comes with unique challenges. Companies often struggle to navigate complex systems, bandwidth limitations, and high storage requirements. One crucial yet overlooked challenge is addressing the bandwidth bottleneck related to 3D depth image compression, an area less highlighted in favor of data rates of evolving RGB-D video streams. This bandwidth crunch can impede the accurate capture of real-world geometry, given the limitations of traditional compression methods in handling abrupt depth discontinuities.
Solution
In our pursuit to solve these complexities, we’ve found a valuable ally in Seeed Studio. Their NVIDIA-based hardware provides the technological backbone we need to bring our solutions to life.
Aivero’s 3D video compression technology provides easy access to annotation, training, and inference tools for 2D and 3D video combinations. Despite how many types of depth cameras you are using, and how many different data formats you accessed from multiple camera SDKs, Aivero simplifies the steps of producing a colorful depth map and converting the data to a point cloud. You can easily preview the point cloud image from specific depth cameras, store the data, and also complete fleet management of the cameras on the visual data capturing platform & dashboard developed by Aivero.
Business Impact/Result
Our 3D video compression technology, empowered by Seeed Studio’s hardware, provides easy access to annotation, training, and inference tools for 2D and 3D video combinations.
In comparison to traditional lossless depth compression methods like RVL, PNG, and LZ4, our solution maintains a high image quality level and operates less computationally expensive. The result? A real-time, low-latency streaming solution that supports up to 3072 depth units or 3.072 meters when using a 1 mm/step resolution. You can store compressed RGB-D video streams to disk and access the compressed data via a suite of languages and tools, including C++, Rust, Python, MATLAB, NVIDIA Deepstream SDK, Samsung NNStreamer, GStreamer, TensorFlow, and MXnet.
Interested in learning more about our compression method and performance metrics? Check out our white paper to compare the most common depth video compression techniques comprehensively.
Strengthening Our Bond with the NVIDIA Jetson Ecosystem
Our partnership with Seeed Studio, an Elite partner in the NVIDIA Partner Network, further strengthens our ties with the NVIDIA Jetson Ecosystem. We invite you to explore more about carrier boards, full system devices, customization services, use cases, and developer tools on Seeed’s NVIDIA Jetson ecosystem page.
