Smart Security Camera With Deep Learning

Wong, Yee Cheng (2020) Smart Security Camera With Deep Learning. Final Year Project, UTAR.

Preview

Abstract

Recent statistics indicate that home burglary remains to be one of the most common property crimes. Video surveillance is one of the mainstream crime prevention measures used around the globe. Driven by the success of machine learning, this study aims to develop a real-time smart security camera system. Different from existing works, the designed deep learning model is deployed on a resource limited Raspberry Pi 3 B+, inference of which is handled by an accelerator called Intel Movidius Neural Compute Stick 2 (NCS2). Darknet-53 of YOLO v3 is selected as the feature extractor and it undergoes transfer learning and fine tuning, in order to detect three types of weapons, namely, gun, helmet, and knife. Furthermore, data augmentation is applied to overcome the scarcity of datasets, which is collected and labelled via LabelImg. The training platform is Google Colab whereas the testing environment using NCS2 requires the pre-trained model to be converted into TensorFlow and subsequently intermediate representation (IR). The performance of crime detection is evaluated in terms of precision, recall, mean average precision (mAP) and frame per second (FPS). Experimental results confirm the effectiveness of using NCS2 in an embedded platform.

Item Type:	Final Year Project / Dissertation / Thesis (Final Year Project)
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions:	Lee Kong Chian Faculty of Engineering and Science > Bachelor of Engineering (Honours) Electrical and Electronic Engineering
Depositing User:	Sg Long Library
Date Deposited:	12 Jun 2021 05:55
Last Modified:	12 Jun 2021 05:55
URI:	http://eprints.utar.edu.my/id/eprint/4046

Actions (login required)

View Item