Machine Learning
NEXUS
ComputerVisionTraffic&EmergencyPre-emption
Problem Statement
Urban traffic congestion leads to increased emissions, delays, and accidents due to inefficient signal timing.
System Architecture
01
Video Feed
02
YOLOv5 Detection
03
ThreadPoolExecutor
04
Docker
05
Signal Control
Solution Overview
Fine-tuned a YOLOv5 detection pipeline for multi-vehicle, emergency-vehicle, and accident detection achieving 81% precision and 80% recall.
YOLOv5 detection pipeline with 81% precision and 80% recall
Multi-vehicle, emergency-vehicle, and accident detection
Parallel video processing using ThreadPoolExecutor
Containerized execution using Docker for reproducible runs
Adaptive signal-control logic with emergency pre-emption
81%
precision
80%
recall
Reduced response delays
reduction
Technology Stack
PythonPyTorchYOLOv5OpenCVDockerMQTTPyQt5AWS