The 6th IEEE International Conference on Artificial Intelligence Circuits and Systems (IEEE AICAS 2024) is being held in Abu Dhabi, UAE from April 22-25, 2024. This conference is a flagship event of IEEE Circuits and Systems Society and one of the most highly regarded annual conferences in this field.
Last year, CenBRAIN Neurotech Center of Excellence was the host organizer of the past edition (IEEE AICAS 2023), Chair Professor Mohamad Sawan, the founding director of CenBRAIN Neurotech, was the general chair of that conference edition.
This year, a paper titled “NeuroSORT: a Neuromorphic Accelerator for Spike-Based Online and Real-Time Tracking” was accepted for presentation and publication in this conference, which led by Ph.D student Ziyang Shen.
第六届IEEE人工智能电路与系统国际会议（IEEE AICAS 2024）正在2024年4月22日至25日，于阿联酋的阿布扎比举办。该会议是IEEE电路与系统领域的旗舰会议之一，也是该领域中最受赞誉的年度会议之一。
去年，先进神经芯片中心担任IEEE AICAS 2023的承办工作，我们的首席科学家，Mohamad Sawan讲席教授担任了该会议的总主席。
今年，一篇名为《NeuroSORT: 基于脉冲的在线实时跟踪神经形态加速器》的论文被IEEE AICAS 2024接受，该文的第一作者是本中心2022级博士生沈子扬。
Abstract
The increasing need for real-time computation with low power consumption is driving the advancement of specialized neuromorphic processors on various applications. Multi object tracking, as one of the most challenging tasks in computer vision, has gained wide attention and many mature solutions with full precision have been proposed. Nevertheless, they consume huge power and computation resources and fail to adapt to edge application scenarios with strict power and resource constraints. A solution with dedicated neuromorphic accelerators for online and real-time object tracking based on spike signals is urgently required. In this work, we propose NeuroSORT, a neuromorphic accelerator for spike-based online and real-time object tracking. NeuroSORT leverages spiking neural network (SNN) to solve linear assignment problem and explores the hardware acceleration on tracking algorithms. Experimental results show that the proposed accelerator reaches an accuracy of 99.43% on linear assignment task and 69.641 HOTA score on MOT17 dataset, while consuming 0.257mW energy and 0.17mm2 area. The overall power consumption is reduced by 41.1% compared with SOTA works with equivalent performance.
低功耗实时计算需求的日益增长，推动了专用神经形态处理器在各种应用领域的发展。多目标跟踪作为计算机视觉领域最具挑战性的任务之一，已受到广泛关注，并有众多研究者提出了许多成熟的全精度解决方案。然而，这些方案消耗巨大的功耗和计算资源，无法适应功耗和资源严格受限的边缘应用场景。基于脉冲信号的在线实时目标跟踪迫切需要一种利用专用神经形态加速器进行加速的解决方案。在这项工作中，我们提出了一种神经形态加速器 NeuroSORT，用于基于脉冲信号的在线实时物体跟踪。NeuroSORT 利用脉冲神经网络（SNN）解决线性分配问题，并进一步探索了跟踪算法的硬件加速。实验结果表明，所提出的加速器搭载相应算法在线性分配任务上达到了 99.43% 的准确率，在 MOT17 数据集上获得了 69.641 的 HOTA 分数，而能耗仅为 0.257mW，所占面积为 0.17mm2。与性能相当的同类型工作相比，总体功耗降低了 41.1%。
Fig 1. A framework of Multi Object Tracking system.
图 1. 多目标跟踪系统框架。
Fig 2. An overview of the proposed TDMPE unit.
图 2. 拟议的 TDMPE 装置概览。
