Design And Development Of Convolutional Neural Network (CNN) Enabled Smart Camera Based User Entrance Intention Detection System

Lai, Ying Chau (2017) Design And Development Of Convolutional Neural Network (CNN) Enabled Smart Camera Based User Entrance Intention Detection System. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text Design And Development Of Convolutional Neural Network (CNN) Enabled Smart Camera Based User Entrance Intention Detection System - Lai Ying Chau - 24 Pages.pdf - Submitted Version Download (400kB)

Abstract

This project aims to design and develop a system of Convolutional Neural Network enabled smart camera to detect user intention of entering a door based on the user’s walking behaviour analysis. Convolutional Neural Network used in this project is 3D Convolutional Neural Network (C3D) integrating with Long Short Term Memory(LSTM). A total of 198 videos, various type of user’s behaviour for walking by with and without entering the access door are collected and added into the training and testing dataset. The collected video resolution is 720P with 29 frames per second and the test image size is 640×360. The dataset is then distributed into 50% for training, 25% for validation and 25% for testing. The videos features extracted from the C3D is used as input to train the Recurrent Neural Network (RNN) that learns to classify video clips of 16 frames. It is first trained to identify walking human and the behaviour of entering a door, entering an access door by scanning an access card and passing by. After clip prediction, the output of the RNN is being post-processed to assign a single activity label to each video, and determine the temporal boundaries of the activity within the video. The experimental results show that the CNN model achieves 99.77% highest accuracy of prediction score and 100% of prediction accuracy.

Actions (login required)

View Item