AI-BASED SMART CLASSROOM CONTACTLESS ATTENDANCE AND LEARNING ANALYTICS PLATFORM

Abstract. To solve low efficiency, error-prone manual statistics and lack of real-time learning data in traditional attendance, this work developed an ai-based smart classroom contactless attendance and learning analytics platform. centered on stm32f1, it integrates k230 vision module for face recognition, uses multi-task cascaded cnn and lbph algorithm, and adds syn6288 speech module. with hardware-software co-design, it works reliably in complex classrooms, realizing automatic contactless attendance and providing data for behavior analysis, offering a cost-effective solution for smart education.

Keywords: smart classroom; contactless attendance; stm32f1; edge computing; embedded system.

Introduction: With the deepening of the Education Informatization 2.0 initiative and the widespread adoption of smart campus development, there is an increasingly urgent need for intelligent upgrades to traditional classroom teaching management models. Currently, low-cost, highly reliable integrated platforms specifically designed for the intelligent management of regular classroom teaching processes remain in the exploratory stage. Existing solutions predominantly rely on high-performance computing units, which result in higher costs and hinder large-scale deployment.

Embedded vision-based solutions represent a significant development direction. However, in classroom environments characterized by complex lighting variations and the requirement for long-term stable operation, achieving real-time, reliable contactless attendance and data collection within limited hardware resources poses a major current challenge. Consequently, this study aims to develop a smart classroom platform centered around the STM32F1 microcontroller, integrated with a lightweight AI vision module. This solution ensures core functionalities while significantly reducing system cost and power consumption, holding substantial practical significance for promoting the equitable access and large-scale deployment of smart education solutions.

I. Technical Design of the Intelligent Platform

1.Hardware System Design

The platform adopts a modular embedded architecture characterized by high integration and low cost. The core controller utilizes a microcontroller from the STM32F1 series, which, with its excellent cost-performance ratio and rich on-chip peripherals, can reliably coordinate the collaborative operation of all system modules. The visual processing unit is handled by an independent K230 intelligent camera module. This module features a built-in NPU (Neural Processing Unit) dedicated to high-speed face detection and feature extraction. It effectively offloads the computational burden from the main controller, thereby resolving the computational insufficiency inherent in a pure software solution based solely on the STM32F1. The voice interaction module employs the SYN6288 Chinese speech synthesis chip, which is controlled via the UART interface of the STM32F1 to achieve targeted voice announcements.

To enhance the system's adaptability to the classroom environment, a basic ambient light sensor is integrated to provide a lighting reference for the image preprocessing algorithms. In terms of communication, the STM32F1 exchanges instructions and data with the K230 module via UART. It also connects to the campus management platform using its built-in USART or an external low-speed Wi-Fi module to achieve scheduled reporting of attendance data.

2.Image Processing and Face Recognition Algorithm

The core face recognition algorithm is fully deployed on the K230 embedded AI module. Within the K230 platform, a lightweight cascaded CNN network is used for rapid face detection, followed by leveraging its NPU hardware acceleration capability for LBPH feature extraction and comparison. This design offloads the computationally intensive vision tasks from the STM32F1 main controller, allowing it to focus on system scheduling, communication control, and logic processing.

The STM32F1 is primarily responsible for sending image capture commands to the K230, and receiving and parsing the structured recognition results it returns. To optimize the workflow, the K230, after completing recognition, transmits only the necessary text or encoded results to the STM32F1 via the serial port. This significantly reduces data transfer volume and the burden on the main controller's resources.

3.System Integration and Communication Control

The system adopts a collaborative "perception-decision-execution" work mode. The K230 module serves as the perception unit, independently performing image capture and face recognition, and packaging the results into custom data frames. The STM32F1 acts as the control hub, receiving the data via the UART asynchronous serial communication protocol, parsing it, and then executing the corresponding operations.

The communication protocol features a streamlined design tailored to the processing capability of the STM32F1, employing a "frame header + data length + command/data + checksum" format to ensure transmission reliability. Based on the recognition results, the STM32F1 drives the SYN6288 module to provide targeted voice feedback. The entire system's business logic is implemented on the STM32F1, with task scheduling based on bare-metal programming or a lightweight RTOS, ensuring the system's real-time performance and stability.

II. Key Implementation Points of the Intelligent Platform

1.Efficient Collaboration under Resource Constraints

Under the resource constraints of the STM32F1, the key to achieving an efficient system lies in clear task partitioning and hardware-software collaboration. High-complexity artificial intelligence algorithms are offloaded to the dedicated K230 module, which functions as a "co-processor." The STM32F1 then focuses on its strengths: real-time control, peripheral management, and handling streamlined communication protocols. The two collaborate through efficient serial communication, thereby enabling the implementation of complex system functionalities as a whole.

2.Ensuring Long-Term Operational Reliability

To meet the demands for long-term continuous operation in classroom scenarios, the system incorporates reliability enhancements at multiple levels. Hardware-wise, industrial-grade STM32F1 chips are selected, and a rational power supply and reset circuit are designed. Software-wise, a hardware watchdog timer is implemented on the STM32F1 to prevent program crashes; a robust communication protocol is designed, including timeout retransmission and data verification mechanisms; and critical data is redundantly stored in the STM32F1's on-chip FLASH or external SPI Flash to prevent data loss. Furthermore, the startup and operational status of the K230 module are monitored to enable abnormal restart, thereby ensuring the overall system's availability.

3.Implementation of Low-Cost Learning Data Acquisition

Regarding learning analytics, this platform focuses on low-cost data acquisition and reporting. While performing face recognition, the K230 module can also calculate simple behavioral indicators, such as "whether seated" and "facial orientation." This preprocessed, low-volume feature information is then sent to the STM32F1. The STM32F1 is responsible for associating this information with timestamps and student IDs to form structured logs, which are cached locally. During breaks or after class, batch data is uploaded to a cloud server via a network module. Complex learning analytics models run on the cloud server, thereby avoiding the high hardware requirements for performing complex computations at the edge. This implements a feasible "edge acquisition, cloud analysis" pathway.

III. Application of Embedded AI Technology in the Education Sector

The trend of cost reduction in embedded artificial intelligence technology is driving intelligent educational devices from the "laboratory" into the "regular classroom." For example, a smart electronic class board based on a similar architecture can handle tasks like class information display and student identity verification. A classroom responder integrated with voice wake-up and recognition modules can achieve low-cost classroom interaction statistics.

The "dedicated AI co-processor + general-purpose microcontroller" architecture adopted by this platform represents a typical current solution for achieving low-cost edge intelligence. It demonstrates how, under constrained resource conditions, advanced face recognition capabilities can be enabled on accessible educational hardware through heterogeneous computing and modular design. This model provides a feasible technical template for the large-scale, high-density deployment of smart classrooms, helping to break down cost barriers and allow intelligent teaching management tools to benefit a wider range of teachers and students.

Conclusion: This study designed and implemented a smart classroom contactless attendance platform based on the STM32F1 microcontroller and an embedded AI module. Through an innovative hardware-software collaborative architecture, this solution successfully integrates high-performance AI visual processing with low-cost, highly reliable system control, achieving the core functionalities on resource-constrained hardware. The system offers the advantages of low cost, high reliability, and ease of deployment and maintenance, providing a practical technical solution for the construction of accessible smart classrooms. Future work will continue to optimize the collaborative efficiency and power consumption between the STM32F1 and K230, and explore networking capabilities with more low-power IoT devices, further promoting the large-scale application of this solution in practical teaching environments.