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

Doull, Katie E.; Chalmers, Carl; Fergus, Paul; Longmore, S.; Piel, Alex K.; Wich, Serge A.

doi:10.3390/s21124074

Cited by 14 publications

(6 citation statements)

References 65 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• layer 1: sixteen convolutional filters (i.e., kernels) with a size of (3, 3), i.e., W 1 has shape (3,3,1,16) This is followed by the ReLU activation function, a strided (2, 2) max-pooling operation and a dropout probability equal to 0.5.…”

Section: Bark Detectionmentioning

confidence: 99%

“…• layer 2: twenty-four convolutional filters with a size of (3, 3), i.e., W 2 has shape (3,3,16,24).…”

Section: Bark Detectionmentioning

confidence: 99%

“…In addition, if a proper network implementation is considered (e.g., with the proper devices, software, and communication protocol) the monitoring of the animal can be performed in real-time to allow exploitation of AAR for various purposes, e.g., study of the interaction between different animals, search and rescue missions [2], protection of animals from poaching and theft, etc. [3]. To perform this, the use of inertial sensors is mandated, such as accelerometers, gyroscopes, and magnetometers as well as a Machine Learning (ML) method, which after the proper training can accurately classify the animal activity [4].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Deep Learning Empowered Wearable-Based Behavior Recognition for Search and Rescue Dogs

Kasnesis

Doulgerakis

Uzunidis

et al. 2022

Sensors

View full text Add to dashboard Cite

Search and Rescue (SaR) dogs are important assets in the hands of first responders, as they have the ability to locate the victim even in cases where the vision and or the sound is limited, due to their inherent talents in olfactory and auditory senses. In this work, we propose a deep-learning-assisted implementation incorporating a wearable device, a base station, a mobile application, and a cloud-based infrastructure that can first monitor in real-time the activity, the audio signals, and the location of a SaR dog, and second, recognize and alert the rescuing team whenever the SaR dog spots a victim. For this purpose, we employed deep Convolutional Neural Networks (CNN) both for the activity recognition and the sound classification, which are trained using data from inertial sensors, such as 3-axial accelerometer and gyroscope and from the wearable’s microphone, respectively. The developed deep learning models were deployed on the wearable device, while the overall proposed implementation was validated in two discrete search and rescue scenarios, managing to successfully spot the victim (i.e., obtained F1-score more than 99%) and inform the rescue team in real-time for both scenarios.

show abstract

Section: Bark Detectionmentioning

confidence: 99%

“…• layer 2: twenty-four convolutional filters with a size of (3, 3), i.e., W 2 has shape (3,3,16,24).…”

Section: Bark Detectionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep Learning Empowered Wearable-Based Behavior Recognition for Search and Rescue Dogs

Kasnesis

Doulgerakis

Uzunidis

et al. 2022

Sensors

View full text Add to dashboard Cite

show abstract

“…The effect of camera angle and time of day on the accuracy of human counts of animals from visual drone images has also received little attention (Augustine & Burchfield, 2022; Brunton et al, 2020; Doull et al, 2021; Linchant et al, 2018; Patterson et al, 2015). Most drone studies conduct animal surveys using a 90° (nadir) camera angle (Borowicz et al, 2018; Dickens et al, 2021; Hodgson et al, 2018; McKellar et al, 2021).…”

Section: Introductionmentioning

confidence: 99%

“…A few studies have compared camera angles for various survey purposes. For example, detection of human poachers in Kenya during visual surveys was greater using a camera angle at 90° than 45° (Doull et al, 2021), but prairie chickens ( Tympanuchus spp.) were best surveyed at leks using camera angles of 10° than 45° and 30°, because it provided a wider angle to see more individuals flushing from the lek (Augustine & Burchfield, 2022).…”

Section: Introductionmentioning

confidence: 99%

Controllable factors affecting accuracy and precision of human identification of animals from drone imagery

Jones,

Elmore,

Krishnan

et al. 2023

Ecosphere

View full text Add to dashboard Cite

Surveying animal populations using drones (unoccupied aircraft systems [UAS]) provides numerous advantages; however, few best practices exist to survey animal communities with drones. Among myriad factors that can affect human identification and counts of animals from drone images, we focused on three factors typically controlled in the study design or by the drone pilot: flight altitude, camera angle, and time of day. Identifying interactions and patterns among these three variables represents an important first step to determining best survey practices. We used a drone to survey known numbers of eight animal decoy species, representing a range of body sizes and colors, at four ground sampling distance (GSD) values (0.35, 0.70, 1.06, and 1.41 cm/pixel) representing equivalent flight altitudes (15.2, 30.5, 45.7, and 61.0 m) at two camera angles (45° and 90°) and across a range of times of day (morning to late afternoon). Expert human observers identified and counted animals from drone images to determine how the three controllable factors affected accuracy and precision. Observer precision was high and unaffected by tested factors. However, results for observer accuracy revealed an interaction among all three controllable factors. Increasing flight altitude resulted in decreased accuracy in animal counts overall; however, accuracy was best at midday than during the morning and afternoon, when decoy and structure shadows were present or more pronounced. Surprisingly, the 45° camera enhanced accuracy more than the 90° camera, but only when animals were most difficult to identify and count, such as at higher flight altitudes or during the early morning and late afternoon. We provide recommendations for improving human accuracy in identifying and counting animals from drone images used in monitoring animal populations and communities. These results should be incorporated into drone survey design in addition to considering funding, logistical, and animal behavior constraints.

show abstract

Monitoring GPS‐collared moose by ground versus drone approaches: efficiency and disturbance effects

Mayer,

Furuhovde,

Nordli

et al. 2024

Wildlife Biology

View full text Add to dashboard Cite

Efficient wildlife management requires precise monitoring methods, for example to estimate population density, reproductive success, and survival. Here, we compared the efficiency of drone (equipped with a RGB camera) and ground approaches to detect and observe GPS‐collared female moose Alces alces and their calves. We also quantified how drone (n = 42) and ground (n = 41) approaches affected moose behavior and space use (n = 24 individuals). The average time used for drone approaches was 17 min compared to 97 min for ground approaches, with drone detection probability being higher (95% of adult female moose and 88% of moose calves) compared to ground approaches (78% of adult females and 82% of calves). Drone detection success increased at lower drone altitudes (50–70 m). Adult female moose left the site in 35% of drone approaches (with > 40% of those moose becoming disturbed once the drone hovered < 50 m above ground) compared to 56% of ground approaches. We failed to find short‐term effects (3 h after approaches) of drone approaches on moose space use, but moose moved > fourfold greater distances and used larger areas after ground approaches (compared to before the approaches had started). Similarly, longer‐term (24 h before and after approaches) space use did not differ between drone approaches compared to days without known disturbance, but moose moved comparatively greater distances during days of ground approaches. In conclusion, we could show that drone approaches were highly efficient to detect adult moose and their calves in the boreal forest, being faster and less disturbing than ground approaches, making them a useful tool to monitor and study wildlife.

show abstract

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

Cited by 14 publications

References 65 publications

Deep Learning Empowered Wearable-Based Behavior Recognition for Search and Rescue Dogs

Deep Learning Empowered Wearable-Based Behavior Recognition for Search and Rescue Dogs

Controllable factors affecting accuracy and precision of human identification of animals from drone imagery

Monitoring GPS‐collared moose by ground versus drone approaches: efficiency and disturbance effects

Contact Info

Product

Resources

About