High-speed industrial machine-vision (MV) applications such as surface inspection of steel sheets necessitate synchronous operation of multiple high-resolution cameras. Synchronization of cameras in the microsecond band is necessary to ensure accurate frame matching while melding images together. Existing approaches for synchronization employ dedicated electronic circuits or network-time-protocol (NTP) whose accuracies are in the millisecond band. Conversely, IEEE-1508 precision-time-protocol (PTP) synchronizes computers in highly accurate industrial measurement and control networks. Synchronization algorithms using PTP involve synchronizing computers connected to cameras. Although the computers synchronize in the microsecond band, the cameras synchronize in the millisecond band. Moreover, PTP is practically not used for synchronizing multiple devices due to the high bandwidth utilization of the network. This paper proposes a temporal synchronization algorithm and framework with two-way communication with timestamps and estimates mean path delays. Unicast transmission forms the basis of the synchronization framework, so that the network utilization is minimal, thereby ensuring the necessary bandwidth is available for image transmission. Experimental results show that the proposed approach outperforms the existing methodologies with synchronization accuracies in the microsecond band.

高速工业机器视觉 (MV) 应用（例如钢板表面检测）需要多个高分辨率相机同步操作。微秒级的相机同步对于确保准确的帧匹配同时将图像融合在一起是必要的。现有的同步方法采用专用电子电路或网络时间协议 (NTP)，其精度在毫秒范围内。相反，IEEE-1508 精确时间协议 (PTP) 在高度精确的工业测量和控制网络中同步计算机。使用 PTP 的同步算法涉及同步连接到相机的计算机。尽管计算机在微秒范围内同步，但相机在毫秒范围内同步。而且，由于网络的高带宽利用率，PTP 实际上不用于同步多个设备。本文提出了一种时间同步算法和框架，具有带有时间戳和估计平均路径延迟的双向通信。单播传输构成了同步框架的基础，使网络利用率最小化，从而确保有必要的带宽用于图像传输。实验结果表明，所提出的方法优于现有方法，同步精度在微秒级。单播传输构成了同步框架的基础，使网络利用率最小化，从而确保有必要的带宽用于图像传输。实验结果表明，所提出的方法优于现有方法，同步精度在微秒级。单播传输构成了同步框架的基础，使网络利用率最小化，从而确保有必要的带宽用于图像传输。实验结果表明，所提出的方法优于现有方法，同步精度在微秒级。