In the realm of energy-efficient technologies, a team of researchers from the University of Southampton, led by Hongyang Shang and Arindam Basu, has made significant strides in optimizing event-based cameras (EBCs) for real-time applications. Their work, published in the IEEE Journal of Solid-State Circuits, focuses on enhancing the performance of corner detection algorithms, which are crucial for tasks like surveillance and autonomous driving.
Event-based cameras, known for their high speed and low power consumption, capture data asynchronously, making them ideal for dynamic environments. However, implementing advanced algorithms like Threshold-Ordinal Surface (TOS) for corner detection on edge devices has been challenging due to high latency. To tackle this issue, the researchers proposed a near-memory architecture called NM-TOS. This architecture leverages a read-write decoupled 8T SRAM cell and employs pipelining to accelerate patch updates. Additionally, it incorporates hardware-software co-optimized peripheral circuits and dynamic voltage and frequency scaling (DVFS) to reduce power consumption and latency.
The NM-TOS architecture demonstrates significant improvements over traditional digital implementations. At a supply voltage of 1.2 V, it achieves a 24.7x reduction in latency and a 1.2x reduction in energy consumption. Even at a lower voltage of 0.6 V, the architecture still shows substantial gains, with a 1.93x reduction in latency and a 6.6x reduction in energy. Monte Carlo simulations confirm the robustness of the circuit, with zero bit error rates at voltages above 0.62 V and minimal errors at lower voltages. The researchers also evaluated the corner detection performance using precision-recall area under curve (AUC) metrics, finding only minor reductions in AUC at 0.6 V for two popular EBC datasets.
The practical applications of this research are vast, particularly in the energy sector. Event-based cameras equipped with efficient corner detection algorithms can enhance surveillance systems, enabling real-time monitoring with lower power consumption. In autonomous driving, these advancements can improve the reliability and efficiency of navigation systems, contributing to safer and more energy-efficient transportation. The NM-TOS architecture’s ability to operate effectively at lower voltages also aligns with the growing demand for energy-efficient technologies, making it a promising development for various industries.
Source: Shang, H., Guo, A., Dong, S., Yang, J., Ke, Y., & Basu, A. (2023). Near-Memory Architecture for Threshold-Ordinal Surface-Based Corner Detection of Event Cameras. IEEE Journal of Solid-State Circuits.
This article is based on research available at arXiv.