IoT AI camera

 Briefly introduce the work: Combining low-power AI chips and mobile networks, the person/object detection model is deployed on the device side to achieve the effects of lower power consumption, real-time response, and traffic saving.

1. Work details: This work implements a camera based on an AI chip that recognizes objects on the device, takes photos, and uploads them to the cloud;

 2. Describe the challenges faced by the work and the problems it solves: This work is aimed at monitoring needs in environments without Ethernet and WIFI, such as wild fish ponds, farms, etc.; in this environment, cellular networks need to be used for data transmission. , if the amount of data is not controlled, high usage fees may be incurred; running the AI ​​recognition model on the device can capture key information while maximizing traffic savings.

 3. Describe the key points involved in the hardware and software parts of the work: 1. The main control chip uses Kanzhi K210, the camera OV7740, and the data transmission uses 4G module EC20 or 2G module SIM800C; 2. The device runs FreeRTOS + LWIP and is driven by PPPOS 4G module, the yolov2-tiny model will be loaded after the system is started, and connected to the back-end server through the MQTT protocol; the camera thread reads the image information, processes it and sends it to the K210's built-in convolutional neural network accelerator to obtain the prediction of yolov2-tiny The results, if there are results exceeding a predetermined classification threshold, are sent to the backend via the MQTT protocol. 3. The back-end server is developed based on Thingsboard to implement image information display and device firmware OTA functions.

4. List of materials for the work: Please refer to the BOM for the main control circuit board. In addition, you also need the OV7740 camera module and 4G/2G module;

5. Upload pictures of works;

* 6. Demonstrate your work and record it as a video for upload; see the attachment for the video