Bias tuning of an event camera for in-cabin automotive environments

This research proposes solutions to the challenges of integrating event cameras into real world applications, with a specific focus on in-cabin driver monitoring systems (DMS). While the world’s first event-based DMS demonstrated the ability to track a driver’s face, count blinks, and estimate head pose and gaze under controlled office lighting, it’s performance was degraded under real-world driving conditions. Factors such as varying lighting environments (e.g., day, night, flickering, and unevenly lit roads) and interference from in-cabin screens or sensors adversely affected performance. To address these challenges, first, we propose novel mathematical metrics, and utilize others including average gradient (AG) and YOLO confidence scores to quantitatively assess DMS performance. Second, a performance monitoring framework was proposed, leveraging these metrics to guide system optimization. Third, a convolutional neural network (CNN) was employed to detect and mitigate specific lighting issue, flickering. Finally, a fully blind optimization approach, using Nelder-Mead algorithm, was implemented to dynamically tune event camera biases (e.g., bias_fo, bias_diff_on, bias_diff_off, bias_refr, and bias_hpf). This approach ensures robust system performance and optimized control of event output without requiring additional filters, thereby minimizing design complexity. The proposed methods significantly enhance event camera performance, improving reliability in challenging scenarios such as low-light environments, high-frequency flickering, and other dynamic conditions. While the focus of this work is primarily on DMS, the findings lay the groundwork for broader applications of event cameras in fields such as robotics, augmented reality, and smart cities. Future work will explore AI-driven adaptive optimization strategies and domain-specific metrics to further enhance the potential of event cameras in next-generation technologies. By bridging the gap between sensor technology and real-world requirements, this research demonstrates that event cameras can transform driver monitoring systems and open up opportunities for advancements in automotive safety, automation, and precision vision systems.

Publisher

University of Galway

Rights

CC BY-NC-ND

cbn

Collections

University of Galway Theses (PhD Theses)