Custom Object Detection with One-stage Detector
Sapkota, Shree (2023)
2023
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-2023060521558
https://urn.fi/URN:NBN:fi:amk-2023060521558
Tiivistelmä
Object detection has become a popular topic of discussion in the field of artificial intelligence (AI). Its use and scope have been widening as the influence of AI technology has been rising. The objective of this study was to develop an object detection model which can be implemented to improve the safety of properties by monitoring unauthorized access.
This study is based on a popular one-stage detector called Yolo (You Only Look Once). In the study, a custom object detection model was created using one of the weights of version 7 of Yolo. A set of image datasets was initially created, which was followed by a data cleaning, annotating and augmenting process. An online platform tool called Roboflow was used in order to create the targeted image dataset. When the dataset was ready, the training was done in Google Colab which uses the Jupyter Notebook environment. Data visualization was done using a Python library called Matplotlib to reflect the performance of the model.
The outcome of this study is a model that is able to identify an unintended human entering a restricted property. This object detection model can be used to secure various public and private properties, for example, in real life situations using a web camera. The next recommended step is enhancing the performance of the model by training more images with higher quality where objects are in different postures, positions, and proximities.
This study is based on a popular one-stage detector called Yolo (You Only Look Once). In the study, a custom object detection model was created using one of the weights of version 7 of Yolo. A set of image datasets was initially created, which was followed by a data cleaning, annotating and augmenting process. An online platform tool called Roboflow was used in order to create the targeted image dataset. When the dataset was ready, the training was done in Google Colab which uses the Jupyter Notebook environment. Data visualization was done using a Python library called Matplotlib to reflect the performance of the model.
The outcome of this study is a model that is able to identify an unintended human entering a restricted property. This object detection model can be used to secure various public and private properties, for example, in real life situations using a web camera. The next recommended step is enhancing the performance of the model by training more images with higher quality where objects are in different postures, positions, and proximities.