During the conduction of this project, the following tasks were performed:
• Reviewed the current state-of-the-art about road safety II concept and authored a Journal paper on this topic
• Reviewed the current state-of-the-art about driving patterns and driver profiles and authored a Journal paper on this topic
• Reviewed the current state-of-the-art about AI applications on Transportation safety and authored a Journal paper and a Conference paper on this topic
• Developed and applied a methodological framework for the identification of driver profiles and authored a Conference paper and a Journal paper (under review) on this topic
• Developed and applied a methodological framework for the recognition of driving patterns and authored a Conference paper and a Journal paper (under review) on this topic
• Developed and applied a methodological framework for the recognition of optimal driving benchmarks and authored a Journal paper (under review) on this topic
• Developed and applied a methodological framework for the recognition of travel patterns and authored a Conference paper on this topic
• Developed and applied a methodological framework to link smooth and safe driving and authored a Conference paper on this topic
• Participation in 5 Conferences and 14 workshops to discuss and disseminate results
• Collaboration with 3 TUD Departments
• Attended 15 training courses
• Developed a website for the Rhapsody project
• Created a YouTube video to describe the project’s methodology to broader audiences
As for the driver profiles, the different driver profiles that exist in the examined sample were identified and are those of the i) less risky drivers, ii) the modest drivers and iii) the more aggressive drivers, which is a finding confirmed by previous literature. It was also found that less risky drivers can be clearly defined by both clustering algorithms especially compared to more aggressive drivers who present a more diverse behavior in terms of driving characteristics. Both algorithms used produced more robust results when the number of clusters was reduced. Finally, no significant association between driver characteristics and clusters was found other than that between clusters and drivers’ age but a larger sample should be collected though to further investigate this.
As for the driving pattern recognition, this research investigated the microscopic characteristics of driving behaviour around driving pulses of 3 types of harsh events of high intensity, namely braking, acceleration and cornering. To this end, a time-series clustering approach was proposed for driving pattern recognition using naturalistic driving data of speed and heart rate for each driver. Results showed that there are 4 to 6 repetitive speed patterns recognized during HB, HA and HC events. The investigation of patterns during harsh braking revealed a steady speed before events, whereas before HA and HC events, a mild to intense deceleration is noticed in almost all patterns. The distribution of observations across patterns of harsh events was not uniform, with some patterns being significantly under-represented. As for the heart rate patterns, it was discovered that these also exist and present upward and downward trends before these events, depending of course on the type of the event.
Regarding the optimal driving benchmark recognition, the boundaries of risky and typical driving were found and those driving periods where quantified for each driver of the sample. For the majority of the drivers recognized as optimal, they appeared not to exceed the speed limit. The rest of the drivers either spent a very short time duration inside the risky period for time to collision (TTC) and crash index (CI) or a longer time duration inside the typical driving period or both. It was highlighted there is no specific optimality pattern that drivers follow but there are several different patterns instead.
