CanMV K230 - (1)

1. Introduction

A variety of CanMV SBCs are manufactured upon Kendryte‘s K230, which is based on th design of T-Head‘s XuanTie C908.

1.1 Online Resource

• K230
• K230 Chip Manual
• K230 OS Images
• K230 IDE

1.2 LCKFB Lushan Pi

Several K230 manufacturers including LCKFB have each designed their own development boards. I purchased the LCKFB Lushan Pi developed and produced by Lichuang (JLC). See the figure below for details:

Clearly, LCKFB Lushan Pi provides 3 MIPI interfaces, and I have 3 GC20X3 MIPI cameras set up for it.

2. Connect LCKFB Lushan Pi and PC Using A Type-C USB Cable

2.1 Flash

For simplicity, I chose CanMV_K230_LCKFB_micropython_v1.6-0-ge4441cc_nncase_v2.9.0.img.gz as the initial firmware image for flashing.

2.2 lsusb

➜  ~ lsusb
...
Bus 009 Device 016: ID 1209:abd1 Generic OpenMV Cam
...

Oh my God… this CanMV is actually using an OpenMV camera module??? (Please refer to my ancient blog OpenMV)

2.3 ll

➜  ~ ll /dev/ttyACM*
crw-rw---- 1 root dialout 166, 0 Apr 20 21:29 /dev/ttyACM0

2.4 From Serial Port

However, it looks it does NOT contain a demo using 3 cameras.

In order to demonstrate the following two examples provided at k230_rtsmart_examples, we need to either flash a different firmware K230 RTOS SDK or build these examples using triple MIPI cameras based on the environment provided by the firmware K230_LCKFB_rtsmart.

• triple_camera_ai: need a display connected to K230 to demonstrate
• triple_camera_ai_rtsp: have to rebuild the source from code

3. Connect LCKFB Lushan Pi and PC Using Serial Cables and A USB-to-TTL Convertor

3.1 Flash

This time, RtSmart-K230_LCKFB_rtsmart_v0.7-0-g0946d74_nncase_v2.9.0.img.gz is flashed onto LCKFB Lushan Pi.

3.2 Cable Connection

Please remember the following 3 points:

• You need a type-C USB cable (the white cable in the picture) used as the power of LCKFB Lushan Pi
• You need a USB-to-TTL (for instance: CH341 or PL2303) for communication via serial port between LCKFB Lushan Pi and your PC
• You need to ensure the RED cable (for power as well) is disconnected. Otherwise, the USB-to-TTL is going to be overheated.

3.3 lsusb

➜  ~ lsusb
......
Bus 009 Device 049: ID 1a86:7523 QinHeng Electronics CH340 serial converter
Bus 009 Device 052: ID 1209:abd1 Generic OpenMV Cam
......

3.4 ll

➜  ~ ll /dev/ttyACM* 
crw-rw---- 1 root dialout 166, 0 Apr 20 21:32 /dev/ttyACM0
➜  ~ ll /dev/ttyUSB*
crw-rw---- 1 root dialout 188, 0 Apr 20 21:32 /dev/ttyUSB0
➜  ~ 

As you can see, in my case, I’m using a CH340.

3.5 List Applications In Serial

Well, this time, we can see triple_camera_ai is already built in the firmware RtSmart-K230_LCKFB_rtsmart_v0.7-0-g0946d74_nncase_v2.9.0.img.gz and flashed into LCKFB Lushan Pi with binary .elf. However, without a display directly connected to my LCKFB Lushan Pi, it’s better to demonstrate with display on a remote PC. Therefore, let’s build triple_camera_ai_rtsp from source.

3.6 Demonstrate triple_camera_ai

4. Build triple_camera_ai_rtsp From Source

4.1 Environmental Configuration

Please refer to K230 RTOS How to Build.

What I’ve done are shown as the following screenshots:

Prepare the source code using repo	k230_rtos_lckfb_defconfig

Deselect CanMV Components Configuration but Select UserSpace Examples Configuration	Enable Auto Detect DRAM Size

Enable Build Triple Camera AI Programs	Entire Process of Configuration

4.2 Build K230 RTOS SDK

4.2.1 Toolchain Preparation

The most up-to-date K230 toolchain can be found on Kendryte Download

4.2.2 Build

The reason why we have to build K230 RTOS SDK is because:

• it generates some files required by those UserSpace examples, including triple_camera_ai_rtsp.
• the built firmware RtSmart-K230_LCKFB_rtsmart_local_nncase_v2.11.0.img is indeed different from the official release RtSmart-K230_LCKFB_rtsmart_v0.7-0-g0946d74_nncase_v2.9.0.img.gz.

4.3 triple_camera_ai_rtsp Demonstration

4.3.1 RTL8152B USB To Wired Converter

4.3.1.1 Find Out IP of Realtek Device Using nmap

4.3.1.2 Network Interface On K230 Using ifconfig

4.3.1.3 Demonstrate MPP Example sample_rtspserver

4.3.4 triple_camera_ai_rtsp Demonstration

Clearly, ISP2 (camera2) detects my face as a single object of interest.