Core Concepts
Leveraging machine learning and computer vision for enhanced safety at Railroad Highway Grade Crossings.
Abstract
The content discusses the critical safety concerns at Railroad Highway Grade Crossings (RHGC) in the United States. It introduces an intelligent system that utilizes machine learning and computer vision techniques to enhance safety at these crossings. The system integrates object detection using YOLO variants and segmentation techniques from the UNet architecture, implemented on a Raspberry Pi. The research achieved high precision rates of 96% for object detection and 98% for segmentation, showcasing the effectiveness of the AI-driven solution. Data collection involved real-time video data from various RHGCs in Nashville, Tennessee, processed for algorithm training. The study also evaluated the performance of the ensemble model and UNet segmentation model through various metrics like confusion matrix, precision-recall curve, training and validation loss, achieving significant accuracy levels.
Stats
The proposed NMS-based ensemble model achieved 96% precision.
The UNet segmentation model obtained a 98% IoU value.
Quotes
"Approximately 95% of all rail-related fatalities over the past decade have resulted from grade crossing collisions or right-of-way trespassing."
"Our research presents a two-pronged computer vision strategy for enhancing safety at RHGCs."