The ESS capture rig accommodates three Canon 6D (or similar) digital SLR cameras. These are controlled from an onboard Raspberry Pi single-board computer running the TriCap data acquisition software.
Consumer DSLRs were preferred over professional aerial survey equipment since the goal was to maximize affordability. Equipment failure, especially mechanical shutter failure, was an initial concern. However, after 6 years of operation and in excess of 700 000 total shutter activations, we have only experienced a single failure.
The rig is designed to be mounted in a SkyReach BushCat Light Sport aircraft (both nose-wheel and taildragger variants), but adapting it to be mounted in a different carriage platform should be perfectly possible.
The BushCat is our preferred carriage platform because it is very affordable and has proven to be extremely reliable in real African bush conditions. It can run on either AvGas or regular 95 octane unleaded petrol, significantly reducing complexity of logistics on remote airstrips where fuel is not always readily available.
The frame accommodates three bodies, each equipped with an 85 mm lens. One camera is pointed straight down, while the other two are tilted to the left and right respectively by 20 degrees each. This arrangement maximizes the width of the imaged strip underneath the plane, while maintaining the viewing angle at 35 degrees from nadir or below. The cameras are controlled via their USB interfaces by a Raspberry Pi single-board computer running custom software that triggers synchronous captures at a programmable frequency. The capture process can be started or stopped using a mechanical switch mounted in the cockpit. The tricap software also presents a web-interface which can be used to adjust capture parameters or view system status.
The cameras have built in GPS receivers, so each image is automatically GPS-tagged in the EXIF metadata. These GPS tags are used to reconstruct the path of the survey aircraft during analysis and to remove any images which were not taken on transect legs (for example during turns) from the data to be processed. The cameras as well as the controlling computer are powered from a Li-ion battery which can power the rig for more than 5 hours.
We have found empirically that elephants can be reliably detected in imagery with a ground sample distance (GSD) of about 10 cm. Our setup allows us to acquire such images from a height of about 4000 feet above ground level (AGL). At this height, the effective search strip width is in excess of 1500 m on the ground. This is dramatically larger than the strip width that can realistically be searched by human observers from any height.
In practice we have found that the neural net performs well on data at this resolution, but human observers (especially inexperienced ones) start to experience significant ambiguity when verifying the results. For this reason, we would often prefer to operate at 3000 feet, reducing both the GSD and the search strip width by a factor 0.75.
Detailed mechanical assembly drawings for the rig are available.