2023

RADIS - Rapid Animal Detection and Identification System for Driver Warning

ML/DL-based Roadside Animal Detection System (RADS) deployed in real-world settings, achieving >97% detection accuracy for wildlife approaching roadways.

Problem Statement

Animal-vehicle collisions pose significant risks to both wildlife and human safety, with millions of animals killed on US roads annually and hundreds of human fatalities. Traditional warning systems are often passive and ineffective.

The challenge was to develop an active, real-time detection system that could identify wildlife approaching roadways and provide immediate warnings to drivers, while operating efficiently in edge computing environments.

Technical Implementation

Hardware Components

  • • Optical camera (SCW "Sharpshooter 2.0", 2MP unit)
  • • NVIDIA Jetson Nano with Maxwell GPU (128 CUDA cores, 472 GFLOPs)
  • • Long-range RF communication infrastructure (XBee emitter/receiver)
  • • Combined wind and solar power module
  • • 4GB LPDDR4 RAM, 16GB eMMC storage, Linux OS

AI Processing Pipeline

  • • YOLO-v5 real-time object detection algorithm
  • • Bounding box creation around moving objects
  • • Classification as vehicle, human, or animal type
  • • Trajectory mapping and speed calculation
  • • Non-Maximal Suppression (NMS) and IoU optimization

Research Methodology

System Design

  • • Low-power edge computing with NVIDIA Jetson Nano
  • • Optical camera system for wildlife detection
  • • RF communication for driver warning signs
  • • Solar and wind power for sustainable operation
  • • Real-time video processing pipeline

AI Implementation

  • • YOLO-v5 computer vision model for object detection
  • • Custom training on feral horse and vehicle datasets
  • • Multi-frame analysis for improved accuracy
  • • Speed and direction calculation algorithms
  • • Automated driver warning system integration

Results & Performance

>97%
Detection Recall
0.91
mAP@0.5 Score
99.16%
System Availability

Field deployment near Reno, Nevada between September 2021 and June 2025 demonstrated the system's real-world effectiveness. The RADIS platform successfully processed thousands of video frames, providing reliable wildlife detection and driver warnings in challenging environmental conditions.

Model Performance

  • • Real-time processing on edge devices
  • • Speed and direction calculation for wildlife tracking
  • • Multi-frame analysis for improved accuracy
  • • Automated driver warning system integration

System Reliability

  • • Zero unforced reboots
  • • No missed heartbeat frames over 10-month period
  • • Automated maintenance cycles (4x daily)
  • • Solar and wind power sustainability

Technologies Used

YOLO-v5
Computer Vision
Edge Computing
NVIDIA Jetson Nano
Python
Linux
RF Communication
Solar Power
OpenCV
CUDA
Real-time Processing
IoT
Embedded Systems

Press Coverage

Conference Presentation

RADIS: Rapid Animal Detection and Identification System for Driver Warning

Authors: Vedant Srinivas, Fraser Shilling

Affiliations: Road Ecology Center, Institute of Transportation Studies, University of California, Davis, USA

Conference: International Conference on Ecology and Transportation (ICOET) 2023

ML/DL-based Roadside Animal Detection System (RADS) deployed in real-world settings, achieving >97% detection accuracy for wildlife approaching roadways. The system combines YOLO-v5 computer vision with edge computing to provide real-time driver warnings and reduce animal-vehicle collisions.

Future Work

Future iterations of RADIS will focus on expanding species detection capabilities, improving power efficiency for longer deployment periods, and integrating with smart city infrastructure for comprehensive traffic safety.

The system architecture is designed to be scalable and adaptable for different geographic regions and wildlife populations, with potential applications in national parks, rural highways, and urban wildlife corridors.