Means and Methods of Efficiency Estimation of Video Stream Transmission Based on GigE Vision Technology Using Application Processor

The paper investigates the possibility of efficient implementation of a GigE Vision compatible video stream source on a computing platform based on a system-on-a-chip with general-purpose ARM processor cores. In particular, to implement the aforementioned video source, a proprietary prototype of a GigE Vision compatible camera was developed based on the Raspberry Pi 4 single-board computer. This computing platform was chosen due to its widespread use and wide community support. The software part of the camera is implemented using the Video4Linux and Aravis libraries. The first library is used for the primary image capturing from a video sensor connected to a single board computer. The second library is intended for forming and transmission of video stream frames compatible with GigE Vision technology over the network. To estimate the delays in the transmission of a video stream over an Ethernet channel, a methodology based on the Precise Time Protocol (PTP) has been proposed and applied. During the experiments, it was found that the software implementation of a GigE Vision compatible camera on single-board computers with general-purpose processor cores is quite promising. Without additional optimization, such an implementation can be successfully used to transmit small frames (with a resolution of up to 640 × 480 pixels), giving a delay less than 10 ms. At the same time, some additional optimizations may be required to transmit larger frames. Namely, a MTU (maximum transmission unit) size value plays the crucial role in latency formation. Thus, to implement a faster camera, it is necessary to select a platform that supports the largest possible MTU (unfortunately, it turned out that it is not possible with Raspberry Pi 4, as it supports relatively small MTU size of up to 2000 bytes). In addition, the image format conversion procedure can noticeably affect the delay. Therefore, it is highly desirable to avoid any frame processing on the transmitter side and, if it is possible, to broadcast raw images. If the conversion of the frame format is necessary, the platform should be chosen so that there are free computing cores on it, which will permit to distribute all necessary frame conversions between these cores using parallelization techniques.