OSCRE: A Jupyter Notebook repository from KennedyClouds

Documentation for the Open Snowflake Camera for Research and Education (OSCRE)

Welcome to OSCRE! This platform is motivated by the need for open, affordable, and upgradeable/adaptable instrumentation to provide microphysical observations for science and educational goals. As the name implies, OSCRE was originally developed to image falling and blowing snow. Components of the system include:

Machine vision camera + lens
LED strobe light
Computer
Off-the-shelf housings mounted to a DIY wood structure
3D-printed parts
Connecting cables

Within this repository, you will find all of the information needed to purchase parts and build OSCRE. Example code to obtain images from the camera are also provided. Contributors are welcomed across all fronts ranging from hardware to software.

Overview of Components and Specifications

Parts, prices, and links to purchase are provided here: Google Sheet

Camera and Lens

Camera

OSCRE uses a singular machine vision camera to obtain images of hydrometeors. These types of cameras are lightweight and rugged, making them ideal for industrial applications such as barcode scanning. A variety of manufacturers exist (Basler, FLIR, JAI, Ximea, etc.) and many of them have models that use identical image sensors from manufacturers such as Sony.

In selecting an appropriate camera/sensor, the following considerations were made:

Affordability (<$1000 USD)
Imaging performance (CMOS, monochrome, global shutter, noise performance)
Minimum exposure time <= 10us (can minimize issues with lighting)
Adequate General Purpose Input/Output (GPIO) control (this provides the signal to the LED light)
Software compatibility with ARM architecture (Nvidia GPU computers)
Adequate frame rate (30+fps)
Connectivity (USB3)

A quick perusal of any of the manufacturers will yield a mind boggling number of choices. The original testbed used a JAI GO-2400M-USB 2.3MP camera, and this is a perfectly acceptable camera for obtaining images. The downside of this camera is a software SDK free for Windows, but requires an expensive (~$500) computer specific license for other architecture. Supposedly this is changing in 2021, but documentation for the 3rd party SDK is also limited.

Instead, development shifted to FLIR for the model documented herein. The FLIR Spinnaker SDK is available for all platforms which means you could use any computer you already own to drive the system assuming the hardware is adequate. Based on the criteria above, the Blackfly S line was selected. Balancing the requirements, the BFS-U3-32S5M-C offered the best bang for the buck. For under $800, the camera uses the Sony IMX252 monochrome sensor to acquire 3.2MP images at rates up to 118fps. The smaller 1/1.8" sensor is advantageous because it offers a longer focal length for a given resolution, increasing the depth of field compared to larger sensors.

Lens

OSCRE uses a Rokinon 135mm f2 lens or 85mm f1.4 lens with manual aperture and focus controls. These are larger lenses designed for traditional cameras. Compared to many (most) machine vision camera lenses, the larger size allows the camera system to only utilize the very center of the optics yielding virtually distortion free images. In other words, a hydrometeor imaged anywhere within the frame will be identical in size. Further, there are no concerns with optical quality (resolution) if the camera is upgraded to a higher resolution sensor in the future. The Rokinon lenses are available in a variety of camera mounts. Both Canon (EF) and Nikon (F) have been used. Note that either lens requires an adapter from the parent mount to the C-mount thread the machine vision camera provides. The 85mm lens is used for cases where you want a smaller form factor platform (focal distance is shorter).

Computer

OSCRE is driven by NVIDIA GPU computers. Currently, the system is designed upon the Jetson Xavier NX Developer Kit which retails for ~$400. Note that as of July 2021, these computers are on backorder due to chipset shortages. It is also possible to drive the camera system with the Jetson Nano, but write performance is not as good due to the lack of a dedicated M.2 slot and slower CPU/GPU. Benchmarking for these platforms is on the to-do list... for educational purposes, the Nano may still be sufficient, as the OS and software will run on either platform. Fall 2022 Update: Best option for the computer is now the reComputer provided by Seeed. The internal computer can be removed from the housing and placed into the 3D printed bracket. Both Nano and Xavier NX kits provide slots for m.2 SSDs! Prices range from $260-$600.

Housings and Structure

OSCRE uses dotworkz S-Type Ring of Fire De-Icing Camera Enclosures (ST-RF-MVP). These housings are waterproof and quite robust... thus far, they have survived two complete winters in North Dakota with no weatherproofing issues. Recently, these housings have been upgraded with stainless steel arms. This means older versions (which are just fine) can be found for discounts on sites such as EBay.