This article comes from the CSK6 visual AI development kit activity organized by Jishu Community and Lens Technology. For more development board trial activities, please pay attention to the Jishu Community website. Author: fat pier
1. Introduction to CSK6 chip
CSK6 is the new generation of AI chip SoC product series of Lens Technology. It adopts multi-core heterogeneous architecture, integrates ARM Star MCU, HiFi4 DSP, and the newly designed AI neural network processing core NPU of Lens, with a computing power of 128 GOPS. The multi-core heterogeneous design enables the chip to meet the AI application requirements of audio, image and video with low power consumption.
Come to Kangkang to see the appearance of the motherboard and the functions of each part of the trial this time, and the visual module is in another part.
The complete board kit looks like this, it's great! There are also network modules (C3 network modules).
2. Development environment installation and practice
The development environment this time is carried out in the environment of window10, and the development process is very smooth. I have to say that this is one of the smoothest development boards I have used. It is very fast to get started, and the tutorials on the official website are also very well written. Follow the documentation step by step to achieve normal results.
1. Install git;
2. Install the CSK6 one-click installation package, the installation package is about 1.8G, and you have to download it patiently;
The CSK6 one-click installation package is an integrated installation package for the CSK6 development environment for the Windows operating system. This installation package will complete a series of operations such as the construction of the CSK6 development environment and SDK deployment, so that you can quickly have an environment that can immediately enter the business development stage.
3. Check the development environment: lisa info zep
4. Create a program: lisa zep create
5. Write the first line of code for "hello_world".
6. Compile the program: first cd to cd hello_worldthe program directory, and then input lisa zep build -b csk6011a_nano
7. Burn the program: lisa zep flash
8. Use the serial port debugging assistant to check the output result.
Main program code:
#include <zephyr/zephyr.h>
void main(void)
{
printk("Hello World! %s\n", CONFIG_SOC);
}
3. Visual Ability and Conception
The original intention was to use the API for face recognition. If the face is correct, it can open the SDK of the door lock. After studying the documents on the official website, I feel that there are still some technical problems that I have not yet thought about. For the time being, I will first experience head and shoulder detection and gesture recognition. They are LIKE( ), OK( ), STOP( ), YES(✌️), SIX( );
1. Install vscode;
2. Install the extension plug-in; csk-application-develop-plugin3.
Download the code lisa zep create --from-git https://cloud.listenai.com/zephyr/applications/app_algo_hsd_sample_for_csk6.git
4. Configure the kit model of the burning program in the vscode plug-in;
5. Open the PC preview function and open the prj.conf file found in the root directory of the project , change CONFIG_WEBUSB=n to CONFIG_WEBUSB=y.
6. Compile the firmware, lisa zep build -b csk6011a_nano
7. Compile the program;
8. Burn the program;
9. Obtain the PC-side test tool. git clone https://cloud.listenai.com/zephyr/applications/csk_view_finder_spd.git
10. Use the Chrome kernel browser to open the index.html file in the project csk_view_finder_spd/src directory;
11. Install the driver, and insert the data cable into the left interface of the development board.
12. Connect the device and start the test.
13. Recognition effect display:
Summary: Next, we will study the powerful functions of this suite in more depth. It is best to apply the network module for data transmission, or store the compressed back-end resources in the device for edge computing. Generally speaking, the recognition rate of the device is relatively good. Except that the photos taken by the single camera are not very clear, there is nothing wrong with it. There is a bold suggestion, I wonder if the vision module can add an infrared camera for live detection.