China
Africa
Explore chapters and articles related to this topic
Real-time driver's drowsiness detection using transfer learning
Published in Sangeeta Jadhav, Rahul Desai, Ashwini Sapkal, Application of Communication Computational Intelligence and Learning, 2022
Mohd Danish Khursheed, Mohd. Maaz Khan, Sara Parveen
Some of the prominent car manufacturers have successfully implemented the technology for driver drowsiness detection systems. The availability of these safety systems is usually limited to luxury vehicles and is not widespread among most drivers. But this task could be achieved by deploying the technique through mobile applications in a much more affordable and convenient manner.
The impact of different seats and whole-body vibration exposures on truck driver vigilance and discomfort
Published in Ergonomics, 2018
Bronson Boi Du, Philip L. Bigelow, Richard P. Wells, Hugh W. Davies, Peter Hall, Peter W. Johnson
A lot research in recent years has focused on the development of early driver drowsiness detection monitors and many of these products are now commercially available. Further research can use these in-vehicle monitors to evaluate future fatigue management interventions (Abe et al. 2014; Lenné and Jacobs 2016). The use of these technologies can help to determine the early decrements in alertness over the course of a truck driver’s shift. In the past, these metrics have been used to identify potential causes of crashes or near misses, (Dingus et al. 2006; Wiegand, Hanowski, and McDonald 2009). The studies were very labour intensive because it required a person to analyse the data frame-by-frame. Now, with technologies available to alert truck drivers when they may be getting fatigued, methods to observe the frequency of lapses are becoming more readily available.
A novel deep learning approach to predict crash severity in adverse weather on rural mountainous freeway
Published in Journal of Transportation Safety & Security, 2023
Md Nasim Khan, Mohamed M. Ahmed
Considering the enormous potential of deep learning in advancing the performance of crash severity prediction models, this study leveraged this cutting-edge technology. Among the various deep learning methods, in recent years, Convolutional Neural Network (CNN) has been widely used in developing prediction models due to its significantly superior performance over other methods, especially with images. Although CNN can be trained from scratches, it requires a huge number of accurately labeled image data, strong computational power, and precise crafting of the model structure. These limitations can be effectively overcome by using pre-trained CNN via a method called transfer learning. The pre-trained CNN models can provide excellent prediction accuracy and can outperform CNN models trained from scratches. Although there are many pre-trained CNNs available in the literature, this study used ResNet18 because of its simple structure, the capability of achieving a high degree of accuracy, and relatively less requirement for computational resources. Unlike other CNNs, ResNet18 uses an optimizable residual learning framework which makes it possible to train deep neural networks with a high degree of accuracy. Considering the robustness of ResNet18, many researchers in the field of traffic safety have applied this pre-trained CNN to develop image-based prediction models. The study by Zhang et al. (2022) applied ResNet18 to identify void defects in airport runways from features collected using a ground-penetrating radar (GPR) system and achieved the highest precision of 90.1%. The use of ResNet18 in detecting adverse weather conditions from road-side webcams was explored in the study of Khan and Ahmed (2022), where the trained model predicted three levels of adverse weather conditions, namely clear, light snow, and heavy snow, with an accuracy of 97%. Another study also used ResNet18 coupled with transfer learning to classify four levels of weather, including sunrise, shine, rain, and cloudy, with an impressive overall accuracy of 98.2% (Al-Haija et al., 2020). By leveraging the features of head and facial expressions, the study by Ma et al. (2021) developed a driver drowsiness detection system based on ResNet-18 and reported an overall accuracy of over 98%. Based on the concept of ResNet18, another study developed a framework capable of detecting, recognizing, and finding boundaries of traffic signs with an accuracy of over 97% (Hrustic et al., 2020).