A cost-effective FWD-RK3588 image processing board, featuring an 8-core 64-bit CPU with a clock frequency of up to 2.4 GHz

This product is an embedded image processing board powered by the Rockchip RK3588 processor, primarily consisting of two components: a core processing board (SoM) and an interface carrier board. The RK3588 is Rockchip’s next-generation flagship high-end processor, fabricated using an advanced 8nm process node. It features an 8-core 64-bit CPU with a primary clock speed of up to 2.4GHz. The processor integrates a built-in AI accelerator NPU, delivering 6 TOPS of computing power and supporting mainstream deep learning frameworks.

The image processing board integrates three video input interfaces, including two MIPI CSI interfaces and one BT.1120 digital video interface, enabling multi-source high-definition video capture. Equipped with abundant onboard communication interfaces, the board achieves stable and efficient video data transmission to meet the demanding requirements of high-performance embedded vision systems. The product has completed core driver adaptation under the Linux (Ubuntu RootFS) environment, allowing users to directly proceed with application-layer development.

This product has completed the adaptation and verification of all core peripheral drivers based on the Linux system (equipped with the Ubuntu file system by default). Users do not need to concern themselves with complex low-level drivers and can directly conduct application-layer development using standard system calls.

• Core Processor Drivers: Adapted the power management, clock, interrupt, and DMA controllers of the RK3588, ensuring efficient scheduling of CPU and NPU computing power.

• Video Input Drivers:

  • MIPI Driver: Adapted the MIPI CSI-2 camera interface and image sensors, supporting camera image data acquisition.

  • BT.1120 Driver: Adapted the digital video receiver chip to enable video data acquisition via the BT.1120 interface.

• Network Interface Driver: Adaptive Gigabit Ethernet, supporting network video stream (RTSP) streaming and remote command control.

• Communication Interface Drivers: Adapted onboard interfaces such as UART, I2C, and GPIO. Users can directly achieve control and data interaction with external devices through file operations or system calls.

The board integrates two types of video input interfaces to meet the connection requirements of different sensors:

• MIPI Digital Video Input: Provides a standard V4L2 interface. Users can acquire image frame data through standard system calls such as open and ioctl.

• BT.1120 Digital Video Input: Provides a unified video device interface to the application layer through the V4L2 framework, facilitating unified software processing.

Integrates a tri-core architecture NPU that supports INT4/INT8/INT16/FP16 hybrid operations, easily handling complex vision tasks such as object detection, classification, and tracking. Provides the RKNN Runtime API, which is responsible for loading RKNN models, managing inputs/outputs, and executing inference computations on the NPU.

To help users get started quickly, we provide running algorithm examples. Users can refer to the implementation logic (note: the examples serve as engineering guidance only; we are not responsible for the performance or efficacy of the algorithms themselves):

• Object Detection: Provides a deployment example of YOLOv5, which can directly perform real-time object detection on the captured video stream and output bounding boxes and categories.

• Object Tracking: Integrates relevant tracking algorithm examples, allowing users to achieve closed-loop control based on miss distance data.

• Integration with Video Streams: The example program demonstrates how to feed image frames into the NPU for inference, and superimpose the inference results (such as bounding boxes) onto the output video via the OSD (On-Screen Display) function, forming a complete end-to-end application closed loop.