Hierarchical Self-Supervised Learning for 3D Tooth Segmentation in Intra-Oral Mesh Scans.

Accurately delineating individual teeth and the gingiva in the three-dimension (3D) intraoral scanned (IOS) mesh data plays a pivotal role in many digital dental applications, e.g., orthodontics. Recent research shows that deep learning based methods can achieve promising results for 3D tooth segmentation, however, most of them rely on high-quality labeled dataset which is usually of small scales as annotating IOS meshes requires intensive human efforts. In this paper, we propose a novel self-supervised learning framework, named STSNet, to boost the performance of 3D tooth segmentation leveraging on large-scale unlabeled IOS data. The framework follows two-stage training, i.e., pre-training and fine-tuning. In pre-training, three hierarchical-level, i.e., point-level, region-level, cross-level, contrastive losses are proposed for unsupervised representation learning on a set of predefined matched points from different augmented views. The pretrained segmentation backbone is further fine-tuned in a supervised manner with a small number of labeled IOS meshes. With the same amount of annotated samples, our method can achieve an mIoU of 89.88%, significantly outperforming the supervised counterparts. The performance gain becomes more remarkable when only a small amount of labeled samples are available. Furthermore, STSNet can achieve better performance with only 40% of the annotated samples as compared to the fully supervised baselines. To the best of our knowledge, we present the first attempt of unsupervised pre-training for 3D tooth segmentation, demonstrating its strong potential in reducing human efforts for annotation and verification.

How do drivers behave during indecision zone maneuvers?

Yellow signal indications at intersections are well-known to be a contributor to traffic crashes. This study examined drivers' behavior during yellow signal indication (i.e., indecision zone) maneuvers. Data from a driving simulator study was used, which included drivers' performance data when they encountered a yellow signal indication at intersections under different secondary-task conditions. This study calculated drivers' likelihood to go through a yellow signal indication and examined factors that are related to drivers' decision making on intersection traversing. The results showed that drivers' decision on stopping or not at a yellow signal indication was associated with different variables including age, distraction, pedal conditions, and time to stop line. Distracted drivers' insensitive behavior was also captured from the significant interaction effect between time to stop line and distraction conditions, which implied that intersection related crash risk may increase when drivers were distracted.

Using naturalistic driving data to examine drivers' seatbelt use behavior: Comparison between teens and adults.

PROBLEM: Teens and young drivers are often reported as one driver group that has significantly lower seatbelt use rates than other age groups. OBJECTIVE: This study was designed to address the questions of whether and how seatbelt-use behavior of novice teen drivers is different from young adult drivers and other adult drivers when driving on real roads. METHOD: Driving data from 148 drivers who participated in two previous naturalistic driving studies were further analyzed. The combined dataset represents 313,500 miles, 37,695 valid trips, and about 9500 h of driving. Drivers did not wear their seatbelts at all during 1284 trips. Two dependent variables were calculated, whether and when drivers used seatbelts during a trip, and analyzed using logistic regression models. RESULTS: Results of this study found significant differences in the likelihood of seatbelt use between novice teen drivers and each of the three adult groups. Novice teen drivers who recently received their driver's licenses were the most likely to use a seatbelt, followed by older drivers, middle-aged drivers, and young drivers. Young drivers were the least likely to use a seatbelt. Older drivers were also more likely to use seatbelts than the other two adult groups. The results also showed that novice teen drivers were more likely to fasten their seatbelts at the beginning of a trip when compared to the other three adult groups. SUMMARY: Novice teen drivers who were still in the first year after obtaining their driver's license were the most conservative seatbelt users, when compared to adult drivers. PRACTICAL APPLICATION: Findings from this study have practical application insights in both developing training programs for novice teen drivers and designing seatbelt reminder and interlock systems to promote seatbelt use in certain driver groups.

Examination of drivers' cell phone use behavior at intersections by using naturalistic driving data.

INTRODUCTION: Many driving simulator studies have shown that cell phone use while driving greatly degraded driving performance. In terms of safety analysis, many factors including drivers, vehicles, and driving situations need to be considered. Controlled or simulated studies cannot always account for the full effects of these factors, especially situational factors such as road condition, traffic density, and weather and lighting conditions. Naturalistic driving by its nature provides a natural and realistic way to examine drivers' behaviors and associated factors for cell phone use while driving. METHOD: In this study, driving speed while using a cell phone (conversation or visual/manual tasks) was compared to two baselines (baseline 1: normal driving condition, which only excludes driving while using a cell phone, baseline 2: driving-only condition, which excludes all types of secondary tasks) when traversing an intersection. RESULTS: The outcomes showed that drivers drove slower when using a cell for both conversation and visual/manual (VM) tasks compared to baseline conditions. With regard to cell phone conversations, drivers were more likely to drive faster during the day time compared to night time driving and drive slower under moderate traffic compared to under sparse traffic situations. With regard to VM tasks, there was a significant interaction between traffic and cell phone use conditions. The maximum speed with VM tasks was significantly lower than that with baseline conditions under sparse traffic conditions. In contrast, the maximum speed with VM tasks was slightly higher than that with baseline driving under dense traffic situations. PRACTICAL APPLICATIONS: This suggests that drivers might self-regulate their behavior based on the driving situations and demand for secondary tasks, which could provide insights on driver distraction guidelines. With the rapid development of in-vehicle technology, the findings in this research could lead the improvement of human-machine interface (HMI) design as well.

Use patterns among early adopters of adaptive cruise control.

OBJECTIVE: The objective of this study was to investigate use patterns among early adopters of adaptive cruise control (ACC). BACKGROUND: Extended use ofACC may influence a driver's behavior in the long-term, which can have unintended safety consequences. METHOD: The authors examined the use of a motion-based simulator by 24 participants (15 males and 9 females). Cluster analysis was performed on drivers' use of ACC and was based on their gap settings, speed settings, number of warnings issued, and ACC disengaged. The data were then examined on the basis of driving performance measures and drivers' subjective responses to trust in ACC, understanding of system operations, and driving styles. Driving performance measures included minimum time headway, adjusted minimum time to collision, and drivers' reaction time to critical events. RESULTS: Three groups of drivers were observed on the basis of risky behavior, moderately risky behavior, and conservative behavior. Drivers in the conservative group stayed farther behind the lead vehicle than did drivers in the other two groups. Risky drivers responded later to critical events and had more ACC warnings issued. CONCLUSION: Safety consequences with ACC may be more prevalent in some driver groups than others. The findings suggest that these safety implications are related to trust in automation, driving styles, understanding of system operations, and personalities. APPLICATION: Potential applications of this research include enhanced design for next-generation ACC systems and countermeasures to improve safe driving with ACC.