TencentOS Tiny AIoT Showcase: Pedestrian Detector, Smart Wheelchair, Wildlife Tracker

At the recently concluded 2021 TencentOS Tiny AIoT Application Innovation Competition attracted more than 50 original works from embedded developers and IoT enthusiasts across the country, covering smart home, animal protection, education, transportation, agriculture and other fields.

Multi‑Function Pedestrian Detector (First Prize)

The award‑winning pedestrian detector is essentially a multi‑purpose target detection device that focuses on pedestrians, offering recognition, counting, and abnormal‑stay behavior detection. It can be used for traffic rule monitoring, pandemic‑related crowd control in malls and scenic spots, and security in unattended areas.

Key highlights include:

Deployment of AI deep‑learning models on embedded boards using cmsis‑nn operators for fast inference with low memory footprint.

The AI model occupies very little memory, running on NXP RT1062, STM32H750, and even low‑power STM32L496 (320 KB RAM) boards.

Support for Tencent LianLian mini‑program to view crowd counts and abnormal behavior.

Smart Wheelchair Application (Second Prize)

This project transforms a traditional wheelchair into an AIoT‑enabled smart wheelchair with remote control, multimodal perception, and cloud connectivity via a Tencent LianLian mini‑program.

The system implements four research areas:

Remote cloud integration with Tencent Cloud and a mini‑program for control.

Smart wheelchair control: foot‑pedal and back‑rest lift, lighting control.

Multimodal perception: internal state (posture, position, speed) and external environment (temperature, humidity, light).

Architecture layers include remote end, main control layer (AIoT development kit with Wi‑Fi), communication layer, perception layer, and driver layer.

Endangered‑Animal Detection & Tracking System (Second Prize)

Using a modified YOLO‑v3 model optimized for MCU deployment, this system detects and tracks endangered wildlife (e.g., Siberian tiger) in the field, reducing storage usage and eliminating manual video review.

Key components:

YOLO‑v3 model trimmed for low‑resolution input and lightweight backbone.

Model deployment on MCU via TensorFlow Lite Micro with quantization.

AIoT integration using ESP8266 to upload detection results to Tencent Cloud.

BMP library to store captured images on MicroSD.

Experts praised the project for replacing manual monitoring with deep‑learning algorithms, greatly improving storage efficiency, lowering costs, and providing a solid foundation for wildlife protection.

The competition also announced popularity and participation awards, and highlighted TencentOS Tiny as a low‑power, modular, secure IoT operating system that enables rapid development of such AIoT applications.