Many remote sensing applications in vegetation science rely on multi-band imaging to construct vegetation indices relating to various eco-physiological processes. Although commercial cameras are available the cost of them runs in the thousands of dollars, while not offering much more than what can be constructed on a more modest budget. Below I describe a simple multi-spectral camera built around a 3D printable housing and off the shelve Raspberry pi and optical components. This puts multip-spectral imaging, for e.g. phenotying, vegetation surveys, crop & environmental monitoring, within reach of those with limited financial means.
Bits and pieces
The camera is built around a Raspberry pi type A or A+, a four camera multiplexer, four 5 or 8 MP Raspberry pi cameras and standard 25mm diameter filters (a common format for bench-top optics). The 3D printed housing can be downloaded from Zenodo (https://doi.org/10.5281/zenodo.4812734). Please cite the work using its DOI when (re)using the design. For custom designs, and software solutions please contact us at firstname.lastname@example.org.
The camera has three main parts, the main housing, a top plate fitting the four cameras and optical filters and cover plate which holds the filters in place. For coloured glass filters a padding ring is provided so the filter is held in place. The top plate and cover plate are screwed into place using M3 screws (using brass threated inserts). The main housing has flanges to provide easy mounting (3 mm mounting holes). A backing plate with a 1/20” camera screw is provided for mounting the camera on a standard tripod. The details of the build are provided below and included in the Zenodo repository.
Below you find some example images of the driveway and surrounding trees. In the current configuration the camera provides an RGB image and spectral bands to calculate both a Normalized Difference Vegetation Index (NDVI) and Photochemical Reflectance Index (PRI). The former is fairly common, while the latter index is harder to interpret but can have added advantages when assessing xanthophyll pigments. Note that the PRI, unlike the NDVI, is poorly defined for non-vegetation areas.
Mount the Raspberry pi A(+) in the housing on the printed stand-offs, using M2.5 screws and bolts. Attach the ribbon cables which are provided with Raspberry pi cameras to the multiplexer board. When mounting the multiplexer board onto Raspberry pi fold the ribbon cables under the board so they all exit at one side of the raspberry pi, inbetween both boards.
Recent batches of the Raspberry pi (NOIR) cameras come mounted with the image sensor on flexible foam double sided tape. This original setup doesn’t allow for easy and predicable alignment of the cameras. To fix this issue, remove the sensor and remount it using thin double sided tape. In some cameras this setup is the default already, if you are lucky you can skip this step.
Mount the cameras using double sided tape, fitting the mounting holes over the printed poles to provide a friction fit. If you are testing a system the tiny poles should provided enough support to keep the cameras in place. For long term applications do use tape as extra support. Be careful while manipulating the cameras as the tiny poles are fragile and can snap if you exert too much pressure.
Carefully connect the ribbon cables to all cameras, routing them appropriately. Push down on the cables gently while closing the top cover, making sure you don’t cover the screw holes.
Finally, install your filters in the filter holes. When using glass filters provide additional padding with a filter ring to prevent the filter from rattling and being damaged (especially on moving platforms). Cover the the filter assembly with the filter plate and screw down firmly.
|filters||2 - 4 (max)|
|filter type||25mm benchtop filters (e.g. Thorlabs)|
|weight||~150gr (as shown)|
|storage||limited by internal SD card, cloud services (wifi) or external USB drive|
Bill of Materials / Cost (as shown)
The cost of the system varies significantly with the quality of the filters used, or the cameras mounted. In the described setup focussing on vegetation indices (EVI and PRI), the laser line filter makes up 25% of the assembly. Using more or less expensive filters will change the price dramatically. The overall price without any filters (or special cameras) comes to 230 EURO (not including 3D printing costs). Beware that hardware modifications for the Rasperry pi cameras exist, in particular there is a monochrome UV capable modification sold by MaxMax (~500 EURO), extending the setup into the UV domain (for direct or fluorescence measurements, or forensic measurements). The header allows for the extension of the setup into the thermal infrared using a FLIR lepton camera unit. Due to current chip shortages I’ve not included this in the current version. The FLIR camera is expensive (~300 EURO) but would increase the spectral domain of the build considerably. For custom builds please contact BlueGreen Labs.
|530nm laser line filter||130|
|570nm bandpass filter||80|
|Raspberry pi A(+)||35|
|Raspberry pi cameras (4x)||140|
|3D printing||free or cheap at a makerspace / institutional workshop|
|threated inserts (M3 / 1/20”)||5|
The Raspberry pi (3) A+ with the faster processor and onboard wifi should probably be avoided for continuous applications. The system is tightly packed and therefore runs hot and overall requires more power than desired (especially for offline use). Using the a Raspberry pi (2) A board avoids these thermal and power issues, while sacrificing internal wifi. Wifi can be added using a USB stick. Future versions might include more venting holes and a smaller (lighter) build around a Raspberry pi Zero. For development purposes I do recommend the Raspberry pi (3) A+ as it makes interacting with the camera faster.