Abstract:<p>The continuous motorization of traffic has led to a sustained increase in the global number of road related fatalities and injuries. To counter this, governments are focusing on enforcing safe and law-abiding behavior in traffic. However, especially in developing countries where the motorcycle is the main form of transportation, there is a lack of comprehensive data on the safety-critical behavioral metric of motorcycle helmet use. This lack of data prohibits targeted enforcement and education campaigns which are crucial for injury prevention. Hence, we have developed an algorithm for the automated registration of motorcycle helmet usage from video data, using a deep learning approach. Based on 91,000 annotated frames of video data, collected at multiple observation sites in 7 cities across the country of Myanmar, we trained our algorithm to detect active motorcycles, the number and position of riders on the motorcycle, as well as their helmet use. An analysis of the algorithm's accuracy on an annotated test data set, and a comparison to available human-registered helmet use data reveals a high accuracy of our approach. Our algorithm registers motorcycle helmet use rates with an accuracy of −4.4% and +2.1% in comparison to a human observer, with minimal training for individual observation sites. Without observation site specific training, the accuracy of helmet use detection decreases slightly, depending on a number of factors. Our approach can be implemented in existing roadside traffic surveillance infrastructure and can facilitate targeted data-driven injury prevention campaigns with real-time speed. Implications of the proposed method, as well as measures that can further improve detection accuracy are discussed.</p>

Motorcyclist helmet detection in single images: a dual-detection framework with multi-head self-attention

UAV Low-Altitude Remote Sensing Inspection System Using a Small Target Detection Network for Helmet Wear Detection

Improved Vision-Based Method for Detection of Unauthorized Intrusion by Construction Sites Workers

Toward Efficient Safety Helmet Detection Based on YoloV5 with Hierarchical Positive Sample Selection and Box Density Filtering

Real-time automatic helmet detection of motorcyclists in urban traffic using improved YOLOv5 detector

Video Based Head Detection and Tracking Surveillance System

Helmet Wearing Detection of Motorcycle Drivers Using Deep Learning Network with Residual Transformer-Spatial Attention

Automatic detection of motorcyclists without helmet

Improved dense residual network with the coordinate and pixel attention mechanisms for helmet detection

Helmet Detection on Motorcyclists Using Image Descriptors and Classifiers

Evaluating Vision-Language Models for Zero-Shot Detection, Classification, and Association of Motorcycles, Passengers, and Helmets

Detecting motorcycle helmet use with deep learning

Dynamic Attention and Bi-directional Fusion for Safety Helmet Wearing Detection

SAFETY HELMET WEARING DETECTION BASED ON AN IMPROVED YOLOV3 SCHEME

DAAM-YOLOV5: A Helmet Detection Algorithm Combined with Dynamic Anchor Box and Attention Mechanism

SHEL5K: An Extended Dataset and Benchmarking for Safety Helmet Detection

Helmet Wearing State Detection Based on Improved Yolov5s

Robust Multiview Multimodal Driver Monitoring System Using Masked Multi-Head Self-Attention

HelmetML: A dataset of helmet images for machine learning applications

Enhanced YOLOv5 algorithm for helmet wearing detection via combining bi-directional feature pyramid, attention mechanism and transfer learning

Object Detection by Attention-Guided Feature Fusion Network