data acquisition engineering

Event-camera perception on highly mobile, resource-constrained mobile platforms faces three fundamental challenges: noise corruption, semantic sparsity, and data deluge. This paper presents a systematic survey of research from 2014 to 2024, offering the first cross-layer analysis—spanning algorithms, models, and hardware—of core bottlenecks and co-design optimization pathways for on-device event data processing. We propose: (i) a lightweight spiking neural network architecture; (ii) a sparse optical flow–driven online filtering method; (iii) heterogeneous computing acceleration strategies leveraging GPU/NPU synergies; and (iv) a multi-sensor fusion framework. Additionally, we curate an open online resource repository and quantitatively characterize real-time–accuracy trade-offs across key tasks—including visual odometry and object tracking. Our work establishes the first reproducible, software–hardware co-optimized deployment roadmap for brain-inspired perception systems on mobile devices.

Heterogeneous camera systems—e.g., visible-light/infrared, professional/consumer-grade, or audio-equipped/audio-less setups—lack hardware synchronization in real-world scenarios, leading to significant spatiotemporal misalignment across multi-view videos. Method: This paper proposes a vision-based time-encoding method leveraging a custom-designed LED Clock. By embedding temporal exposure timestamps within frames using red and infrared LEDs, the approach achieves cross-modal, audio-free, and external-timecode-free millisecond-level synchronization. It further integrates RMSE-optimized temporal alignment with joint multi-device calibration. Contribution/Results: The method reduces synchronization residuals to 1.34 ms—substantially outperforming existing optical signal, audio-based, and timecode synchronization schemes. Validated in large-scale surgical recordings involving over 25 heterogeneous cameras, it significantly improves downstream tasks including multi-view 3D reconstruction and pose estimation.

This work addresses the limited flexibility in algorithm development and method comparison inherent in existing wearable eye-tracking systems by presenting a low-cost, modular, and extensible binocular eye-tracking platform. Built from off-the-shelf components and 3D-printed structures, the system integrates four infrared eye cameras, infrared illumination, and an optional scene camera, accompanied by calibration and synchronized data acquisition software. Designed to prioritize research adaptability over end-user robustness, the platform supports multiple eye-tracking paradigms—including stereo, glint-based, and binocular approaches—within a single hardware configuration. Feasibility is demonstrated through a functional prototype, and all hardware designs and documentation are openly released to facilitate reproducibility and further innovation.

This work addresses the challenge of time synchronization among heterogeneous sensors in roadside and vehicle-mounted multi-LiDAR–multi-camera systems by proposing an open-source, modular, and scalable hardware synchronization solution. Using the LiDAR synchronization pulse as a reference, the system employs programmable delay circuits to generate independent trigger signals for each camera, enabling flexible and precise spatiotemporal alignment. The architecture supports arbitrary combinations of sensor counts and has been validated on both a three-camera roadside platform and a seven-camera vehicular setup. Experimental results demonstrate significantly improved spatial consistency between point clouds and images, while the design ensures robustness, reproducibility, and ease of deployment.

Temporal misalignment among multi-event cameras arises from trigger and transmission delays; hardware synchronization is constrained by circuit dependencies and device incompatibility (e.g., CeleX5 lacks support). Method: We propose a hardware-agnostic software synchronization approach that leverages the onset-time disparity of event density distributions across cameras. Our method jointly models event density, optimizes distribution similarity, and dynamically adjusts timestamps for high-precision temporal alignment. Contribution/Results: This framework overcomes hardware limitations, significantly enhancing system compatibility and deployment flexibility. Extensive experiments across diverse scenes and camera models (e.g., DAVIS346, CeleX6) demonstrate consistently sub-10 ms synchronization error, enabling accurate fusion of multi-view event data for downstream tasks such as 3D reconstruction and motion estimation.

Event cameras face challenges in real-time, pixel-level flicker fundamental-frequency estimation under high temporal resolution and wide dynamic range, constrained by full-sensor readout bandwidth. To address this, we propose an asynchronous, per-pixel online frequency imaging method. Our approach comprises: (1) a lightweight luminance approximation reconstruction using second-order IIR filtering; (2) a high-precision period estimation algorithm combining falling-edge triggering with zero-crossing interpolation; and (3) the first lightweight ROS node supporting 50 Meps throughput, enabling full-sensor real-time processing on a single CPU core. The method achieves a per-pixel fundamental-frequency detection上限 of 64 kHz with accuracy comparable to Prophesee’s proprietary module. An open-source implementation has been validated on embedded near-sensor computing platforms.

This work proposes a fully fixed-point, non-iterative, streaming optical flow algorithm tailored for efficient deployment on resource-constrained FPGAs. By partitioning asynchronous event streams into fixed-time windows and representing them as 1-bit spatial occupancy grids, the method evaluates multiple velocity hypotheses in parallel using only integer logic—comprising shift registers, counters, comparators, and LUT-based multipliers—without requiring frame reconstruction, floating-point arithmetic, or division operations. A single-axis prototype was successfully implemented on a Xilinx Artix-7 FPGA, occupying less than 2 kB of memory and achieving 99.5% directional accuracy under event densities of 10–40%. To the best of our knowledge, this is the first demonstration of low-latency, sparse velocity estimation on an FPGA without relying on DSP blocks or dedicated dividers.

This work addresses the computational inefficiency of motion parameter estimation in event-based vision systems, which hinders real-time performance on embedded platforms. The paper presents the first FPGA-specific hardware architecture tailored for contrast maximization algorithms, integrating event warping, contrast computation, and a hardware-aware iterative solver through a deeply pipelined design. This enables high-throughput, low-power processing of asynchronous event streams. Compared to conventional CPU/GPU implementations, the proposed architecture accelerates motion parameter estimation by over two orders of magnitude, demonstrating significant improvements in both real-time capability and energy efficiency, as validated in event-based object tracking tasks.

This work addresses the challenge of identifying periodic artificial targets from the asynchronous, sparse event streams generated by event cameras. To this end, the authors propose the FRIES framework, which employs temporal gating to suppress background noise, combines pixel-wise activity maps with region clustering to generate regions of interest, and performs local spectral analysis on each region to extract dominant frequencies. Furthermore, a Resonant Time Surface (RTS) is introduced to enable frequency-selective visualization, enhancing events whose phases align with the target frequency. This study pioneers the application of frequency-domain analysis to event-based vision, leveraging the high temporal resolution of event streams for periodicity-driven target discrimination and selective monitoring. Experiments demonstrate accurate recovery of mechanical chopper and drone rotor frequencies indoors, as well as effective detection of hovering drones against complex outdoor foliage backgrounds.

Traditional spaceborne optical sensors are hindered by motion blur, high power consumption, and data redundancy, limiting their efficacy for efficient Earth observation. This work systematically reviews recent advances in event-based cameras for spaceborne remote sensing and introduces, for the first time, a four-dimensional classification framework tailored to space applications—encompassing high-speed atmospheric observation, environmental change detection, on-orbit intelligent processing, and geospatial modeling—to underscore the transformative potential of event vision as a new paradigm in remote sensing. Leveraging inherent characteristics such as asynchronous sensing, neuromorphic engineering, high dynamic range imaging, and microsecond-level temporal resolution, this technology substantially enhances observational efficiency while reducing both energy consumption and data volume, thereby offering a promising pathway toward sustainable space exploration.

This work addresses the challenge of irreversible loss of task-critical information under bandwidth, latency, and power constraints inherent in conventional downsampling approaches. The authors propose the first task-driven, real-time foveated imaging system that dynamically allocates pixel budgets at capture time through a dual-stream sensor architecture: high-resolution sampling is focused on task-relevant regions while low-resolution captures preserve global context. Formulating gaze control as a sensor attention policy learning problem, they employ reinforcement learning to enable closed-loop optimization between perception and acquisition. This approach overcomes the limitations of static or non-task-aware capture strategies, significantly outperforming existing baselines under strict pixel budgets and demonstrating practical feasibility on a 200-megapixel dual-stream sensor for real-world video acquisition.