MYIR RK3576 Board: 12-Channel 1080p HD Video Streaming

In today's era of rapidly advancing intelligent vision technology, the efficient processing and transmission of multi-camera data have become a core requirement in various application scenarios. Whether enabling comprehensive monitoring in smart security systems, supporting precise detection in industrial vision gateways, facilitating real-time assistance through vehicle-mounted surround-view platforms, or ensuring rapid response in AI-driven community security systems, high-efficiency and low-latency solutions remain fundamental to turning these applications into reality.

Current multi-channel video transmission solutions often face a number of limitations, such as a restricted number of input channels that fall short of the requirements of large-scale surveillance scenarios. These systems are also plagued by high latency, which undermines real-time performance, along with subpar encoding efficiency that typically leads to excessive bandwidth and storage consumption. However, MYIR has successfully achieved efficient H.264 encoding and low-latency RTSP streaming for 12-channel HD video streams on the Rockchip RK3576 SOM, with an end-to-end latency of approximately 140ms. In the following section, we will elaborate on how MYIR accomplishes low-latency end-to-end transmission for 12 high-definition video streams.

1. The powerful performance architecture of the MYD-LR3576 development board

The MYD-LR3576 development board is based on the Rockchip RK3576 SoC, which utilizes an advanced 8nm process and integrates a quad-core Cortex-A72 (clocked at 2.2GHz) and a quad-core Cortex-A53 (clocked at 2.0GHz). It is equipped with a Mali-G52 MC3 graphics processing unit, a VPU video codec unit, and features a 6TOPS computing power NPU.

MYD-LR3576 Development Board based on Rockchip RK3567

2. The 12th camera captures the video stream, encodes it and pushes it via RTSP. On the receiving end, the stream is decoded and displayed.

The video demonstrates that the RK3576 board captures video input from twelve cameras at 1920x1080 resolution and 30 frames per second. The encoded video streams are transmitted over a network cable to another RK3576 board for decoding and display. A dedicated camera focused on the stopwatch is used to accurately measure the end-to-end delay.

STEP 1: Use 12 AHD cameras to capture the stream, encode it, and push the stream

The MYD-LR3576 development board features three 4-lane MIPI-CSI interfaces. When paired with the MY-CAM004M video conversion module, it can connect up to twelve AHD high-definition cameras (1920x1080@30fps). The video streams are processed via RGA for image optimization,

then encoded into H.264 format by the VPU unit, and finally pushed via RTSP over the network. The

entire process achieves an end-to-end latency of less than 80–100 ms.

The workflow of capturing video streams, encoding, and pushing via RTSP:

RGA processing unit occupancy

CPU processor unit occupancy

STEP 2: MYD-LR3576 Development Board: Video Stream Reception and Display

CPU occupancy 

GPU occupancy 

Due to network transmission and decoding processes, there will be a delay of approximately 40ms to 50ms. Consequently, the total end-to-end delay, encompassing video capture, encoding, RTSP transmission, decoding, and playback, falls between 120ms and 150 ms.

Compare the occupancy of each unit in the two processes

3. Extensive application scenarios

In the context of industrial security, handling 12 camera streams is little more than a warm-up for the RK3576. It's important to note that the chip also features a 6 TOPS NPU, which has not yet been fully utilized. With this capability, the platform is well-equipped to support the following application scenarios:

Intelligent Security Monitoring: As a central hub, it can easily control the video stream from multiple surveillance cameras to realize comprehensive, real-time monitoring without any blind spots.

Industrial Vision Processing Gateway: accurately process all kinds of vision data in industrial production to facilitate intelligent manufacturing.

Surround-View Fusion Platform: seamlessly integrates feeds from multiple vehicle cameras, enhancing driving safety with reliable and efficient visual assistance.

AI-Powered Security for Smart Communities: Monitors (falling objects), detects strangers, and prevents e-bikes in elevators, etc.

In summary, MYIR’s SOM solution, developed on Rockchip’s RK3576 SoC, has successfully overcome key challenges in multi-channel HD video integration, high-efficiency encoding, and low-latency transmission. This technological advancement offers strong support for real-time performance and system reliability in intelligent vision applications. Soon, we will share a detailed implementation guide on AI-powered multi-video stream processing using the MYIR RK3576 SOM. Stay tuned!