View on GitHub

Computer Vision and Aerial Imagery for Wildlife Conservation

Collection of recent works on using Computer Vision (CV) and aerial imagery for wildlife conservation

This survey is intended to consolidate recent efforts on animal conservation using remote imaging (UAVs, manned flights, or satellite imagery). This page is maintained by Benjamin Kellenberger and Dan Morris. Contributions are welcome! Please submit additions or corrections as GitHub pull requests.

1. Contents

1. Contents
2. Research Articles
3. Case Studies
4. Further Reading

2. Research Articles

Papers are ordered alphabetically by the last name of the first author. If possible, links, images and a brief summary are provided.

Corcoran, Evangeline, et al. Automated detection of koalas using low-level aerial surveillance and machine learning. Scientific reports 9.1 (2019): 3208.

The authors employ an ensemble of off-the-shelf Faster R-CNN and YOLO models (predictions fused by averaging) to detect koalas. Noteworthy detail: double-counts due to overlapping images are resolved by finding correspondences (via ORB features) and averaging the detection heatmaps in the overlap zones.

Dataset available on request.

Keywords: deep learning, faster r-cnn, wild mammals

      

van Gemert, Jan C., et al. Nature conservation drones for automatic localization and counting of animals. European Conference on Computer Vision. Springer, Cham, 2014.

Traditional ML methods (DPM, eSVM) for cattle detection. They also do counting using videos and a KLT tracker. One of the early works, with a straightforward dataset.

Keywords: SVM, tracking, cattle

      

Hong, Suk-Ju, et al. Application of Deep-Learning Methods to Bird Detection Using Unmanned Aerial Vehicle Imagery. Sensors 19.7 (2019): 1651.

Comparison of standard detectors (Faster R-CNN, R-FCN, SSD, YOLO) on birds. Interesting detail: they also use bird decoys for data augmentation.

Keywords: deep learning, faster r-cnn, r-fcn, ssd, yolo, birds

      

Kellenberger, Benjamin, Diego Marcos, and Devis Tuia. Detecting mammals in UAV images: Best practices to address a substantially imbalanced dataset with deep learning. Remote sensing of environment 216 (2018): 139-153.

This paper presents tricks to handle the scarcity of animals in UAV imagery.

Dataset freely available.

Keywords: deep learning, resnet, wild mammals

      

Okafor, Emmanuel, et al. Operational data augmentation in classifying single aerial images of animals.”2017 IEEE International Conference on INnovations in Intelligent SysTems and Applications (INISTA). IEEE, 2017.

Okafor, Emmanuel, Lambert Schomaker, and Marco A. Wiering. An analysis of rotation matrix and colour constancy data augmentation in classifying images of animals. Journal of Information and Telecommunication 2.4 (2018): 465-491.

These papers focus on maximizing the variability of the animals a CNN gets to see; the authors position individuals in all rotational variations across the screen, in-painting holes with nearest neighbors.*

Keywords: deep learning, cattle

      

Rey, Nicolas, et al. “Detecting animals in African Savanna with UAVs and the crowds.” Remote sensing of environment 200 (2017): 341-351.

Mammal detection in the African savanna using hand-crafted features (histogram of colors, bag of visual words) and an exemplar SVM.

Keywords: exemplar SVM, wild mammals

      

Rivas, Alberto, et al. Detection of Cattle Using Drones and Convolutional Neural Networks. Sensors 18.7 (2018): 2048.

Heatmap-based animal detection in low-altitude imagery that provides relatively clear foreground/background separability. The authors use a custom CNN that is similar to LeNet.

Keywords: deep learning, cattle

      

Andrew William, Colin Greatwood, and Tilo Burghardt. Visual localisation and individual identification of Holstein Friesian cattle via deep learning. Proceedings of the IEEE International Conference on Computer Vision. 2017.

Combination of off-the-shelf models (R-CNN and a KCF tracker) for cattle monitoring.

Keywords: deep learning, kcf, tracking, cattle

      

Longmore, S. N., et al. Adapting astronomical source detection software to help detect animals in thermal images obtained by unmanned aerial systems. International Journal of Remote Sensing 38.8-10 (2017): 2623-2638.

This paper does peak finding by thresholding, followed by HOG and SVM, to detect animals in thermal imagery.

Keywords: svm, thermal, cattle

      

Bondi, Elizabeth, et al. Spot poachers in action: Augmenting conservation drones with automatic detection in near real time. Thirty-Second AAAI Conference on Artificial Intelligence. 2018.

An integrated, cloud-based framework that uses Faster R-CNN on thermal data to localize poachers.

Keywords: deep learning, faster r-cnn, thermal, poaching

      

Bondi, Elizabeth, et al. AirSim-W: A Simulation Environment for Wildlife Conservation with UAVs. Proceedings of the 1st ACM SIGCAS Conference on Computing and Sustainable Societies. ACM, 2018.

Uses a rendering engine (AirSim, based on the Unreal Engine) to simulate an African ecosystem and derive thermal signatures for detector training.

Keywords: deep learning, thermal, simulation

      

3. Case Studies

Delair / Bathawk

Collaboration between drone manufacturer Delair and Tanzanian anti-poaching service Bathawk Recon. Used mainly IR footage for detecting poachers and mammals. Presented as a video.

      

Mara Elephant Project / DJI

Collaboration between the Mara Elephant Project and drone manufacturer DJI. Presented as a video.

      

Where’s Bunga? Tracking Orangutans with Drones and Gnu-Radio

Technical POC created by Dirk Gorissen in collaboration with International Animal Rescue. Deployed a semi-automatic drone to identify orangutans in the Bornean rainforest.

https://dirkgorissen.com/2019/01/06/wheres-pinoh-tracking-orangutans-with-drones-and-gnu-radio

https://dirkgorissen.com/2016/04/19/wheres-susi-airborne-orangutan-tracking-with-python-and-react-js

      

JGI / ConservationDrones.org

Collaboration between the Jane Goodall Institute of Canada and ConservationDrones.org to explore drone use in a Congolese reserve. Mentioned applications include censuses of chimpanzees. Presented as a video.

https://conservationdrones.org/2015/05/06/view-from-above-using-drones-to-help-support-jane-goodall-institute-conservation-efforts-in-congo/

      

Deep Reality Simulation for Automated Poacher Detection

Leverages AirSim-W (see paper by Bondi et al. above) to generate synthetic thermal imagery for training poacher detection models. Uses Faster R-CNN, domain adaptation using GANs, and an implementation using MS Azure and Apache Spark.

https://databricks.com/session/distributed-deep-domain-adaptation-for-automated-poacher-detection

Also see accompanying video.

      

Hensoldt

German sensor manufacturer Hensoldt presents a program intended to stem rhino poaching in South Africa in this video.

      

4. Further Reading

ConservationDrones.org blog: applications (incredible collection of additional case studies)

ConservationDrones.org blog: publications (incredible collection of additional research papers)

Awesome Deep Ecology (review of deep learning applications in ecology)

Camera Trap ML Survey (analogous list for camera traps)