Object detection for general object type using mobilenet-SSD neural network database

Lai, Jiun Hong (2021) Object detection for general object type using mobilenet-SSD neural network database. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Object detection for general object type using mobilenet-SSD neural network database.pdf - Submitted Version Download (1MB)
	Text (Full Text) Object detection for general object type using mobilenet-SSD neural network database.pdf - Submitted Version Restricted to Repository staff only Download (6MB)

Abstract

Object detection has been experienced a dramatically technological improvement in the computer vision area. Besides that, the object detection technique is the combination of object classification and object localisation. The main task for object detection is to classify the object categories and identify the object's position in a certain image or video. Recently, advances in deep learning have significantly enhanced the efficiency of object detection techniques in speed and precision. This has allowed modern desktop computer systems to conduct extremely effective object detection at the real-time level. Other than that, the advancement in creating the high performance of deep neural network architectures for the embedded system has been increasing rapidly. Besides that, this project aims to investigate the appropriateness of operating object detection on the Raspberry Pi 4 Model B. Next, Raspberry Pi also is known as a popular embedded computer board. By comparing the performance of the SSD algorithm and MobileNet-SSD algorithm, especially on the accuracy, there is one experimental framework that should be carried out by using the different parameters, which is the size of input images (280 x 280 pixels, 320 x 320 pixels and 360 x 360 pixels) and the range distance between camera and object (0.6 meters until 1 meter). Furthermore, five types of objects will undergo in the experiment: a computer mouse, cell phone, remote, laptop, and keyboard. Based on the overall result obtained, it can be concluded that the MobileNet-SSD algorithm has better performance, which has an almost 76% average detection rate compared with the SSD algorithm, which only has a 65% of average detection rate in the detection system. Hence, the MobileNet-SSD algorithm is more suitable for running real-time object detection on Raspberry Pi with high accuracy and precision than the SSD algorithm. Apart from that, Raspberry Pi is suitable to use as hardware in a real-time object detection system. It can replace the traditional desktop system to reduce the maintenance fee and indirectly improve work efficiency.

Actions (login required)

View Item