Intelligent Object Recognition for Temperature-Based Fan Control
Wong, Yun Sang (2024)
2024
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2024112831089
https://urn.fi/URN:NBN:fi:amk-2024112831089
Tiivistelmä
This thesis presents the development and implementation of intelligent object recognition on a temperature-based fan control system, aimed at advancing automation through more thoughtful and adaptive solutions. The proposed system is designed to recognize cats and humans using a machine learning model deployed on an edge device, contributing to smart home technology, human and pet care products, and environmental control systems focusing on sustainability and energy efficiency.
The project involved data collection, labelling, model training, and deployment on Raspberry Pi 3, along with the design of a temperature-based fan control system, a sensor platform printed circuit board design, 3D enclosure design. The image data for the model training was captured using a smartphone camera with labelling performed in Labellmg. Model training was conducted on Google Colab, using a pre-trained SSD-MobileNet-v2 (FP32) TensorFlow model. The trained model achieved a mean average precision (mAP) score of 80.83% across various Intersections over Union (IoU) thresholds and 97.55% at a 0.5 IoU threshold. The model was then deployed on Raspberry Pi 3. The temperature-based fan control system, which includes a DHT11 sensor, a DC fan, and an LCD, successfully adjusted fan speed based on user input and temperature data.
As a result, the integrated system successfully detected objects which were cats and humans, performing tasks based on the detected objects and temperature status.
The system’s performance met expectations.
The project involved data collection, labelling, model training, and deployment on Raspberry Pi 3, along with the design of a temperature-based fan control system, a sensor platform printed circuit board design, 3D enclosure design. The image data for the model training was captured using a smartphone camera with labelling performed in Labellmg. Model training was conducted on Google Colab, using a pre-trained SSD-MobileNet-v2 (FP32) TensorFlow model. The trained model achieved a mean average precision (mAP) score of 80.83% across various Intersections over Union (IoU) thresholds and 97.55% at a 0.5 IoU threshold. The model was then deployed on Raspberry Pi 3. The temperature-based fan control system, which includes a DHT11 sensor, a DC fan, and an LCD, successfully adjusted fan speed based on user input and temperature data.
As a result, the integrated system successfully detected objects which were cats and humans, performing tasks based on the detected objects and temperature status.
The system’s performance met expectations.