Enhancing Road Safety- Telematics Data Analysis and Driver Profiling — CU Boulder Capstone Reflection
Written by: Srimedha Bhavani Chandoo, Prathik Bharath Jain, Uday Gadge, Siddharth Tayi, Jagrati Chauhan, Brahmendra Charan Attanti
In today’s fast-paced world, where mobility is a cornerstone of daily life, the protection and performance of transportation systems have become very critical. Driver profiling, a concept born out of the intersection of technology and transportation, offers a promising avenue for understanding and improving driver behavior on roads. Driver profiling, at its core, entails the analysis of various factors influencing driver behavior to ascertain individualized driving scores. These scores serve as a metric for evaluating driving habits, identifying areas for improvement, and ultimately fostering safer road conditions.
The project’s primary focus revolves around integrating factors such as speed, sudden movements in a single trip, and lane changing into the profiling process. The aim is to enhance our comprehension of human behavior behind the wheel and its implications for road safety by enhancing the telematics data provided. This initiative represents a commitment to sustainable mobility and is being led by 99P Labs.
Understanding the Data:
Telematics and V2X Data Telematics is like the digital heartbeat of modern vehicles. It’s the technology that links cars and trucks to a vast network of information, turning them into smart, data-driven machines. Imagine your vehicle not just as a mode of transportation but as a hub of communication, constantly exchanging vital data with other vehicles, infrastructure, and even with the driver. At its core, telematics collects data from various sensors embedded in vehicles. These sensors monitor everything from engine performance and fuel consumption to location, speed, and sometimes even driver behavior. This data is then transmitted wirelessly to central servers for analysis and action.
Telematics empowers vehicle owners, fleet managers, and automotive companies with real-time insights into vehicle health, usage patterns, and operational efficiency. V2X, or Vehicle-to-Everything communication, takes telematics to the next level. It’s the technology that allows vehicles to communicate not just with each other (V2V) but also with roadside infrastructure (V2I), pedestrians (V2P), and the cloud (V2C). This two-way communication opens up a world of possibilities for enhancing road safety, traffic management, and overall driving experience.
With the use of cutting-edge telematics technology, we are able to access an immense amount of V2X data that has been gathered from automobiles in multiple locations for this project. This data provides a complete picture of the activity and health of the cars, not just their location. We have access to information such as location, speed, braking patterns, fuel usage, engine diagnostics, and tire pressure. The dataset’s schema comprises several tables, including Summary Table, Host Table, Spat Table, RvBSM, and evtWarn. The Summary and Host Tables measure and report vehicle data, while the Spat Table stores data from smart junctions interacting with cars. The RvBSM table involves interactions between remote cars and smart intersections, stored within host cars. Lastly, the evtWarn table captures data reported by remote cars, providing valuable insights into potential warnings or alerts. The data contains 6193 trips of 157 drivers.
Data Enhancement and Analysis:
- Lanes change and overtaking behavior
In analyzing lane change and overtaking behavior using V2X data, we delved into the critical relationship between yaw rate, speed, and time during these maneuvers. By examining temporal variations in yaw rate before, during, and after maneuvers, we gained insights into anticipatory behavior, execution proficiency, and potential risks associated with abrupt lane changes and overtaking events. Utilizing latitude and longitude coordinates, we visualized trip routes, facilitating detailed driving behavior analysis and route optimization. This analysis identifies overtaking events based on intervals between yaw rate peaks, highlighting instances where vehicles successfully change lanes and overtake others. However, the increase in overtaking, especially with rapid speed changes, raises safety concerns and emphasizes the need for responsible driving practices to mitigate accident risks.
2. Speeding and Break Patterns
Analyzing the speed dynamics during lane changes revealed critical insights into driver behavior and safety. We investigated the correlation between speed and lane change execution across various phases, from approach to integration. Excessive speeds during lane changes increase collision risks due to reduced maneuverability and compromised vehicle control, while overly cautious speed adjustments can impede traffic flow. Moreover, our examination of brake application patterns during lane changes highlights the importance of proactive risk management. Abrupt or erratic braking behaviors signal potential risks and inadequate situational awareness, emphasizing the need for consistent and well-timed brake utilization for smooth lane change execution and overall road safety.
3. Sudden acceleration and deceleration
Driver profiling through the analysis of sudden acceleration and deceleration events is a crucial aspect of improving vehicle safety and performance. This analysis revealed driver behavior, vehicle functionality, and overall road safety dynamics, highlighting aggressive driving behaviors like speeding, tailgating, and abrupt braking. Our approach involves identifying harsh driving maneuvers within each trip assigning penalty scores based on percentiles and initially using thresholds at the 97.5th and 2.5th percentiles. We refined our methodology to include additional thresholds at the 95th and 5th percentiles and the 99th and 1st percentiles. This nuanced scoring system accurately reflects the severity of driving behavior, ensuring appropriate penalties for varying levels of abrupt movements. By aggregating penalty scores across trips, we gain a comprehensive understanding of driving behavior and safety performance, enabling data-driven strategies for promoting responsible driving practices and enhancing overall road safety standards.
4. Integrating Speed Limits
This project focused on integrating speed limits into the V2X dataset to assess driver adherence to traffic rules and safety. Leveraging location data (latitude and longitude) updated every 0.5 seconds, we used the OpenStreetMap API as a cost-effective alternative to obtain road information and corresponding speed limits. Challenges included mapping data volume and accurately locating roads instead of landmarks. The solution involved smoothing speed vs. time curves using the Savitzky-Golay filter and selecting local maxima points, significantly reducing data points while retaining trip information. The API provided 70% of speed limit values, with the remainder filled through regression-based imputation using driver speed, trip speed limit averages, and driver speed limit averages. This approach ensures sufficient data for analysis while highlighting areas for future improvement.
5. Analyzing turns behavior
The analysis of turn behavior involved several key steps. First, distinguishing turns from other maneuvers like lane changes or curved roads. Next, calculate the total number of turns per trip and plot their distribution to identify potentially risky driving instances. This includes scoring turns based on their frequency and duration. Additionally, analyzing maximum yaw rate values helps differentiate between turns, lane changes, and U-turns. Yaw rates between 5 and 35 are typically associated with turns, while values below 5 indicate lane changes or curves, and values over 35 may indicate U-turns. Further analysis involves calculating the average speed and time taken for each turn partition to detect rash driving behaviors such as high-speed turns in a short duration. Overall, this approach offers insights into driver behavior during turns while aiding in identifying potentially unsafe driving practices.
6. Clustering Analysis
The clustering analysis in this project aimed to understand driver behavior patterns using interpretable features like speed and acceleration. Initially, we grouped drivers into three categories based on their speed habits: those within speed limits, consistently slower drivers, and drivers that frequently exceed the speed limits. Expanding our analysis to include vehicle dynamics and engine data provided deeper insights, establishing thresholds for various features and validating driving behaviors through hypothesis testing. This approach not only allowed for nuanced driver profiling but also paved the way for future segmentation based on a wider range of driving characteristics. Additionally, we conducted hypothesis testing to compare individual driver speeds against the overall average, crucial for understanding how different driving styles can impact traffic flow.
Results and Findings:
In the analysis of the V2X dataset for January 2022, a substantial portion of the trips provided valuable speed-related information, with a high success rate in locating road names and speed limit data. The integration of a regression model to predict missing speed limits further enhanced the dataset’s completeness, achieving impressive accuracy metrics. This approach not only enriched the dataset but also facilitated insights into driver behavior regarding speed compliance, contributing significantly to the understanding of driver safety and behavior patterns.
The clustering and sudden acceleration and deceleration analysis delved deeper into driver behavior, revealing distinct profiles among three driver groups. Cluster 0 exhibited law-abiding tendencies with moderate brake usage and consistent adherence to turn signals. Conversely, Cluster 1, characterized by cautious driving, showed heightened brake usage and lane change frequency, reflecting a safety-oriented approach. Cluster 2, representing speed enthusiasts, displayed lower stability metrics and minimal brake involvement, indicating a more aggressive driving style.
The included image provides a visual representation of the driver clustering process. Each data point represents a trip from a particular device, and its color indicates the cluster (0, 1, or 2) it belongs to based on its speed behavior. This initial clustering was performed on individual trips
We then took an additional step. We aggregated the trip data for each device and performed clustering again using the same features. The results were remarkably consistent! Most devices remained assigned to the same cluster as their trips, suggesting a stable driving style across their journeys. This consistency highlights the effectiveness of our clustering approach in capturing inherent driver behavior patterns. Devices 10419,10032,10429 exhibit the same characteristics.
These findings, consistent across individual trips and device-level data, underscored the robustness of the clustering methodology in capturing nuanced driver behaviors and preferences.
Future Directions: Innovating Road Safety Strategies
The comprehensive analysis of driver behavior in this project sheds light on critical aspects such as lane changes, overtaking maneuvers, and sudden accelerations/decelerations. Moving forward, there is an opportunity to delve deeper into predictive modeling using machine learning algorithms. By leveraging advanced analytics techniques and real-time data streams, we can develop predictive models that anticipate risky driving behaviors, enabling proactive interventions to prevent accidents.
Furthermore, the integration of additional data sources, such as weather conditions and road infrastructure data, can enhance the sophistication of our safety monitoring systems. Incorporating environmental factors into our analysis will provide a more holistic understanding of driver behavior and its interaction with external variables. This holistic approach, coupled with continuous research and refinement, will pave the way for innovative strategies aimed at fostering safer road environments and promoting responsible driving practices for the benefit of all road users.
Thank you 99P Labs!
The project team expresses deep gratitude towards 99P Labs for granting us the opportunity to engage in a captivating analysis within this dynamic problem domain, leveraging their invaluable data resources. We sincerely appreciate the collaborative efforts and support extended by 99P Labs throughout our capstone project journey, which has enriched our exploration and insights into this intriguing field.
