Video analysis for the detection of animals using convolutional neural networks and consumer-grade drones

Chalmers, Carl; Fergus, Paul; Montañez, Casimiro Aday Curbelo; Longmore, Steven N.; Wich, Serge A.

doi:10.1139/juvs-2020-0018

Cited by 18 publications

(12 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is encouraging that AI for video classification is fundamentally no different to that for photo classification in that each video frame can be treated like a single photo (e.g. Chalmers et al., 2019), and that AI video analysis offers other potential advantages over photo classification (e.g. Johanns et al., 2022).…”

Section: Discussionmentioning

confidence: 99%

Camera trapping with photos and videos: implications for ecology and citizen science

Green

Stephens

Whittingham

et al. 2022

Remote Sens Ecol Conserv

View full text Add to dashboard Cite

Camera traps are increasingly used in wildlife monitoring and citizen science to address an array of ecological questions on a wide variety of species. However, despite the ability of modern camera traps to capture high-quality video, the majority of studies collect still images, in part because of concerns with video performance. We conducted a camera trap survey of a forested landscape in the UK, using a grid of paired camera traps, to quantify the impact of using video compared to photos on the outcomes of ecological research and for participation and engagement of citizen scientists. Ecological outputs showed no difference between photo and video datasets, but comparison between expert and citizen science classifications showed citizen scientists were able to classify videos more accurately (average accuracy of 95% for video, 86% for photo). Furthermore, citizen scientists were more likely to volunteer additional information on age (provided for 61% videos and 30% photos) and sex (provided for 63% videos and 45% photos) of animals in video footage. Concerns over slow trigger speeds for videos did not appear to affect our datasets or the inferences gained. When combined with citizen science, video datasets are likely to be of higher quality due to increased classification accuracy. Consequently, we encourage researchers to consider the use of video for future camera-trapping projects.

show abstract

Section: Discussionmentioning

confidence: 99%

Camera trapping with photos and videos: implications for ecology and citizen science

Green

Stephens

Whittingham

et al. 2022

Remote Sens Ecol Conserv

View full text Add to dashboard Cite

show abstract

“…), both TIR and RGB images, via the framework: www.conservationai.co.uk (accessed 9 April 2020) using the Visual Object Tagging Tool (VoTT) version 1.7.0. In order to classify objects within new images, the deep neural network extracts and 'learns' various parameters from these labelled images [28,38].…”

Section: Automated Detection Softwarementioning

confidence: 99%

“…Therefore, a downside of this method is that the training and the use of these models in near real-time are extremely costly, along with a steep learning curve, due to the complex network and computational requirements. However, research is currently being conducted to overcome these challenges [38,60].…”

Section: Manual Vs Automated Detectionmentioning

confidence: 99%

“…A strategy called 'You Only Look Once' (YOLO) uses a lightweight model with a single CNN, and although this provides faster inference time, it has the downside of having a poor overall accuracy of detection. This accuracy worsens when the drone is flown at higher altitudes [38,71]. More advanced drones and software can be used to overcome these limitations, having longer operating times and higher spatial resolutions, but they are restrictive in terms of their cost, which is unrealistic for small NGOs in developing countries.…”

Section: Challenges For the Usage Of Drones In Conservation Managementmentioning

confidence: 99%

See 1 more Smart Citation

An Evaluation of the Factors Affecting ‘Poacher’ Detection with Drones and the Efficacy of Machine-Learning for Detection

Doull¹,

Chalmers

Fergus

et al. 2021

Sensors

Self Cite

View full text Add to dashboard Cite

Drones are being increasingly used in conservation to tackle the illegal poaching of animals. An important aspect of using drones for this purpose is establishing the technological and the environmental factors that increase the chances of success when detecting poachers. Recent studies focused on investigating these factors, and this research builds upon this as well as exploring the efficacy of machine-learning for automated detection. In an experimental setting with voluntary test subjects, various factors were tested for their effect on detection probability: camera type (visible spectrum, RGB, and thermal infrared, TIR), time of day, camera angle, canopy density, and walking/stationary test subjects. The drone footage was analysed both manually by volunteers and through automated detection software. A generalised linear model with a logit link function was used to statistically analyse the data for both types of analysis. The findings concluded that using a TIR camera improved detection probability, particularly at dawn and with a 90° camera angle. An oblique angle was more effective during RGB flights, and walking/stationary test subjects did not influence detection with both cameras. Probability of detection decreased with increasing vegetation cover. Machine-learning software had a successful detection probability of 0.558, however, it produced nearly five times more false positives than manual analysis. Manual analysis, however, produced 2.5 times more false negatives than automated detection. Despite manual analysis producing more true positive detections than automated detection in this study, the automated software gives promising, successful results, and the advantages of automated methods over manual analysis make it a promising tool with the potential to be successfully incorporated into anti-poaching strategies.

show abstract

“…Although research estimating population density from drone surveys is emerging (e.g., Beaver et al 2020), the vast majority of the drone-based studies thus far have focused on determining species' presence (Linchant et al 2015;Wich and Koh 2018;Wang et al 2019). Drones can survey areas in a fraction of the time of other existing methods (Jiménez López and Mulero-Pázmány 2019), and observer bias (i.e., differences between observers in their ability to detect the presence of the animal of interest) can be minimized as multiple observers can review images or video footage obtained (Vermeulen et al 2013;Martin et al 2015;Scarpa and Piña 2019) and machine learning algorithms can be used to automatically detect species or individuals (Seymour et al 2017;Corcoran et al 2019Corcoran et al , 2020Chalmers et al, 2021). In addition, a wide variety of sensors (e.g., multispectral or hyperspectral imaging outside of the typical RBG frequency range, LiDAR, chemical imaging) can be mounted on drones to achieve particular desired research objectives (Wich and Koh 2018;Jiménez López and Mulero-Pázmány 2019).…”

Section: Introductionmentioning

confidence: 99%

Detecting spider monkeys from the sky using a high-definition RGB camera: a rapid-assessment survey method?

et al. 2022

View full text Add to dashboard Cite

show abstract

Video analysis for the detection of animals using convolutional neural networks and consumer-grade drones

Abstract: xxx

Cited by 18 publications

References 26 publications

Camera trapping with photos and videos: implications for ecology and citizen science

Camera trapping with photos and videos: implications for ecology and citizen science

An Evaluation of the Factors Affecting ‘Poacher’ Detection with Drones and the Efficacy of Machine-Learning for Detection

Detecting spider monkeys from the sky using a high-definition RGB camera: a rapid-assessment survey method?

Contact Info

Product

Resources

About