discussion / Biologging  / 23 September 2018

WildFind - a package to produce a georeferenced heatmap of VHF collared animals.

Wildfind

This work was initiated after experiencing ‘greater than normal’ frustrations in identifying the location of the VHF-collared animals at a particularly remote field site which had complex topography, vegetation and a complete absence of light aircraft.  It was around this time that I saw a video that a colleague had posted on Youtube showing First Person View (FPV) operated aircraft.

https://www.youtube.com/watch?v=kjkVYxoUwxc&t=

So, the germ of an idea began.  Could I utilise this FPV technology to detect radio-collared animals? Turns out, the answer was yes.  With little background in electronics, I built a system that demonstrated proof of concept.

https://youtu.be/TjxsnbCagUg

Yep, Lego.  (This was a hobby project)

This VHF receiver was too heavy so I switched to a Communication Specialists R1000 receiver. In the meantime I learnt how to build a quadcopter.  We flew a short mission that was sufficient to demonstrate the directionality of the transmitter could be detected. This was a little surprising given the electrical noise that was on-board and around the vehicle. The VHF receiver was still far too heavy.

https://youtu.be/myYyq3SZ2f8


The next critical step was learning about Software Defined Radio dongles – more information at this link.  So my project then became focussed on use of these devices as the VHF receiver. Initially, running SDR# on a laptop but then switching to a Raspberry pi board computer.

https://youtu.be/WyexyhFR9Sc

Along the way, I observed that others had been thinking along similar lines, (head nods to Gary Mortimer in South Africa and also Lian Pin Koh & Serge Wich of Conservation Drones).

Conceptually, the UAV would search a large area – essentially similar to an aerial photogrammetry mission – I was overlooking the requirement for Beyond Line of Sight UAV operation certification at this stage.

There is some value in building a system that simply uses the aircraft to fly to 120 m / 400 ft altitude immediately above the pilot, i.e. effectively building an elevated listening post, but I thought that the system would be more powerful if it could search a much larger area and report the general location and status (alive / dead) of the animal.  This is the basic information that I typically require on a daily basis in my field studies. 

A heat map of signal strength would be the ideal output of a successful system – utilising a low gain omni-directional antenna. Use of a high-gain system would lead to much larger ‘hot’ zones in the heat map and frustrate localisation of the animal. This was now getting well beyond my hobby skill level and I contacted Al Brown (https://eartoearoak.com ) with a request for advice.  He responded in a very positive manner and the project matured. Contributions from numerous other people were also forthcoming and welcome  – thanks to Tom Kaminski and Hansi Wegner.

Al coded the WildFind package and various versions were tested along the way. This was somewhat complicated by the fact that Al was based in the UK, Tom in USA and Hansi and I in Australia. Most testing was undertaken using ground-based vehicles on a laptop computer as it just simplified the process – but the package was intended to run on an Odroid XU-4 board computer suitable for use in the UAV. All of my testing was using relatively cheap SDR dongles from RTL-SDR.com.  

On one occasion, I was able to put my laptop, SDR dongle and the low quality rigid omni antenna into a helicopter that was undertaking a low-level grid-pattern flight over one of my field sites where there were three radio-collared feral goats. This flight pattern was pretty much the intended operation for Wildfind vehicle.

The output of this flight was ‘interesting’ (see attached screenshot) – but did not accurately reflect the reality of where the goats were known to be. The package reports the frequencies of the collars that it detects, and the collar transmission rate. When the animal is alive and moving, the collars typically transmit the pulsed tone at 40 ppm and this switches to 80 ppm when the animal ceases movement (after a 10 hour delay for example).

The tables to the right of the heatmap image show 17 different collars where there was only three animals known to be in the site. It is not known whether the errors are related to problems with Wildfind code, hardware errors or electrical noise associated with the helicopter operation.

 

One of the main difficulties we think is the caused by the short burst transmission of the VHF collar that leads to difficulties for WildFind to receive, analyse and accurately report the collar. So currently, Sept 2018, there is still some way to go before the system is polished.

Al has also investigated the use of an artificial intelligence solution but found that this was prohibitively power hungry for use on UAV.  

Ideas are welcome – the code is hosted here -   https://github.com/EarToEarOak/Wild-Find

 

Mike Johnston

 

 

 

 

 

 




Hi Mike,

I spoke with Al only a couple of weeks ago and he seemed to have shelved the project as he couldn't take it any further using current hardware. He stated that the software ran incredibly slow even on desktop computers. I do see the usefulness in having something like this as a conservation tool for guys on the ground, in the air and within drones. I am interested in seeing this being picked up and developed further, perhaps with higher grade hardware. Shame as it is a great concept and good to see that someone else values a tool such as this. 

Hello Mike

If I understood your system description correctly, you intend to analyse the radio data and produce the heatmap on board the drone. Why ?

I would do minimal data processing onboard, log any relevant details onto  removable memory such as an SD card and do all further processing offline on a PC.

This should dramatically reduce the energy and processing demands of your drone-based hardware and simplify the software, too.

I guess there must be (open source?) solutions available that can convert data in a suitable format into a heatmap. This should save you having to re-invent that wheel.

Hope this is of some use.

Good luck,

Joachim Neff

 

 

 

 

 

 

Hello Mike

I am not sure whether a datalogger would resolve radio noise problems. I would expect more success from improving the antenna/receiver circuitry to get rid of the noise before you digitise/record. Unfortunately, I can't give you much advice on this subject but antenna/circuit design can be tricky.

I just watched your YouTube clips and noticed in the last clip that there seems to be a 40 dB margin between the transmitter signal and the background noise. Based on my limited experience, this should be plenty to clearly separate signal from noise. With the right type of signal conditioning/filters you should be able to log only the signal you are interested in. Record this together with the current location coordinates and you should get close to the data set you need for producing the heatmap.

Good luck,

Joachim Neff