AI Wildlife Detection and Road Hazard Prevention
An edge AI wildlife detection system designed to identify Australian native animals in real time and alert drivers before potential road collisions.
Background
Wildlife collisions are a serious and ongoing hazard across Australia's regional, rural and urban roads. Kangaroos, wallabies, wombats and other native animals frequently enter roads at dawn, dusk and during low-visibility conditions, giving drivers little time to react. This project delivers an end-to-end hardware and software system covering data collection, labelling, model training, edge deployment, monitoring and real-time driver alerting.
Challenges
- No production-ready system existed for detecting Australian native wildlife on roads in real time
- A large and representative image dataset was required across different animals, lighting conditions, terrain and road environments
- The machine learning model needed to operate efficiently on compact edge hardware
- Inference needed to be fast enough for real-time driver alerting
- The system needed to remain operational in remote areas with limited or unavailable cloud connectivity
Our Solution
- Collected and labelled tens of thousands of Australian wildlife images using Label Studio
- Stored and managed the training dataset through AWS S3
- Trained a YOLOv8 object-detection model using AWS SageMaker
- Optimised the model for Australian wildlife classes including kangaroos, wallabies, wombats and other road-hazard animals
- Deployed the trained model onto Jetson Nano edge hardware running Ubuntu
- Designed an alerting and monitoring layer for notifying drivers when wildlife is detected
- Enabled the deployed system to operate independently of continuous cloud connectivity
Results
- Tens of thousands of training images labelled
- Real-time object-detection capability
- Jetson Nano edge deployment
- Offline-ready inference
- Remote-road deployment architecture
10s of 1000s
Training images labelled
Jetson Nano
Edge deployment
Real-time
Driver alerting
Offline-ready
Remote-road operation
Collision-reduction and measured safety outcomes have not been claimed. This system has not yet been evaluated in production.
Technology Stack
Related Services