A Deep Learning Model for Global Camera Trap Labelling
Lead Research Organisation:
Brunel University London
Department Name: Inst of Environment, Health & Societies
Abstract
Recent years have seen an increase in camera-trap survey monitoring by ecological researchers. Camera-trap surveys collect imagery of medium-large mammal species across a region of interest. Dependent on the activity in an area and the number of camera trap days a survey's conducted, millions of images may be captured. Currently, researchers label each image with species and behavioural information or enlist Citizen Science volunteers to assist in doing so.
Developments in machine learning, specifically Deep Learning, has provided promising image recognition methods and may be utilised to assist in the camera trap labelling.
The research aims to produce a technique that allows ecological researchers to go from raw images through animal detection, species classification, and analysis identifying species behaviours based on the underlying predictions, all as an automated process.
Developments in machine learning, specifically Deep Learning, has provided promising image recognition methods and may be utilised to assist in the camera trap labelling.
The research aims to produce a technique that allows ecological researchers to go from raw images through animal detection, species classification, and analysis identifying species behaviours based on the underlying predictions, all as an automated process.
People |
ORCID iD |
Allan Tucker (Primary Supervisor) | |
Benjamin Evans (Student) |
Publications
Norman D
(2022)
Can CNN-based species classification generalise across variation in habitat within a camera trap survey?
in Methods in Ecology and Evolution
Description | Developed methods to improve the accuracy of animal detection in camera trap imagery and to speed up the labelling of camera trap imagery for ecologists. |
Exploitation Route | Assist ecologists in faster labelling and analysing the growing camera trap surveys utilised to learn about species occupancy and behavioural dynamics. |
Sectors | Digital/Communication/Information Technologies (including Software) Education Environment |
Title | CamTrap Detector |
Description | Cross Platform (Windows, macOS, Linux) Graphical Interface for detecting Animals, Humans and Vehicles in Camera Trap Imagery with options for various export formats. |
Type Of Technology | Webtool/Application |
Year Produced | 2022 |
Open Source License? | Yes |
Impact | Making the MegaDetector models accessible to those outside of the machine learning and programming communities. Utilised by the HogWatch team at Zoological Society London. |
URL | https://github.com/bencevans/camtrap-detector#readme |
Title | CamTrapML |
Description | Python Library containing Detection and Utility functions for working with Camera Trap Surveys |
Type Of Technology | Webtool/Application |
Year Produced | 2022 |
Open Source License? | Yes |
Impact | Eased the usability of MegaDetector models from Python programs. |
URL | https://github.com/bencevans/camtrapml |