Helmet Use Detection of Tracked Motorcycles Using CNN-Based Multi-Task Learning

Abstract

Automated detection of motorcycle helmet use through video surveillance can facilitate efficient education and enforcement campaigns that increase road safety. However, existing detection approaches have a number of shortcomings, such as the inabilities to track individual motorcycles through multiple frames, or to distinguish drivers from passengers in helmet use. Furthermore, datasets used to develop approaches are limited in terms of traffic environments and traffic density variations. In this paper, we propose a CNN-based multi-task learning (MTL) method for identifying and tracking individual motorcycles, and register rider specific helmet use. We further release the HELMET dataset, which includes 91,000 annotated frames of 10,006 individual motorcycles from 12 observation sites in Myanmar. Along with the dataset, we introduce an evaluation metric for helmet use and rider detection accuracy, which can be used as a benchmark for evaluating future detection approaches. We show that the use of MTL for concurrent visual similarity learning and helmet use classification improves the efficiency of our approach compared to earlier studies, allowing a processing speed of more than 8 FPS on consumer hardware, and a weighted average F-measure of 67.3% for detecting the number of riders and helmet use of tracked motorcycles. Our work demonstrates the capability of deep learning as a highly accurate and resource efficient approach to collect critical road safety related data.

Positional Encoding: Improving Class-Imbalanced Motorcycle Helmet use Classification

“…In this method, each class is regarded as a specific combination of rider position and individual riders' helmet use, as a result, the model suffers from imbalanced class distribution. Multitask learning [15] simultaneously predicts helmet use class and visual similarity of a given image pairs. Using a reweighting process, the Class-balanced (CB) loss method [20] is applied to three widely used loss functions, Softmax crossentropy (CE) loss, Sigmoid CE loss, and focal loss.…”

“…The HELMET dataset [15] was used to evaluate our proposed model. It contains 910 video clips recorded from 12 observation sites in Myanmar, where each video clip has 100 frames.…”

“…For comparison purpose, we used the same data splitting as [15]. The dataset is split into three non-overlapping subsets, i.e., training set (70%, 636 video clips), validation set (10%, 92 video clips), and test set (20%, 182 video clips).…”

“…To allow road safety actors and regulators to continuously monitor motorcycle helmet use, a number of computer vision based approaches for the automated detection of helmet use have been proposed [12] [13]. In addition, due to the significant relation between helmet use and the position of riders on the motorcycle, more recent studies have implemented measures to differentiate between driver and passenger helmet use through the application of deep learning [14] [15].…”

Why so serious? - Comparing two traffic conflict techniques for assessing encounters in shared space

“…Unlike other naturalistic observation methods, such as field operational tests (FOT) or naturalistic driving studies (NDS), data collection is not done for individual road users only, but an observation space is filmed to investigate interactions of all road users at this particular section of the road environment. But while extensive advancements have been made in the computer vision enhanced analysis of road user behaviour, a number of shortcomings remain (Artan, Bulan, Loce, & Paul, 2014;Lin, Deng, Albers, & Siebert, 2020;. In this study, a relatively high number of encounters had to be discarded, due to interruptions of the detection or difficulties in the discrimination between different road users.…”