Brain Tumor Classification

Brain tumor is a group of abnormal cells in the brain which can be either cancerous (malignant) or non-cancerous (benign). Gliomas are the most common brain tumor that can be categorized into low-grade glioma (LGG) and high-grade glioma (HGG). Identifying the location and area of the tumor has been a tedious job for radiologists, let alone to classify the tumor into the LGG / HGG. Hence, a fully automatic approach is proposed to segment the brain tumor in Magnetic Resonance Imaging (MRI) images and classify it into the LGG / HGG. A high bias algorithm 3D Convolutional Neural-network (CNN) is employed to segment the brain tumor into edema, necrotic, and enhancing tumor. The segmented tumor is then used to generate the radiomics features and the Support Vector Machine (SVM) is used to classify the LGG and HGG tumor. This work is tested on Brats 2017 and the experiments show promising results with the Dice scores, for segmentation and classification for LGG and HGG, at 0.85 Dice Scores and 0.83 F1-Score, respectively. View Project

Crack Detection System

Reinforced concretes and cement based materials are used to construct buildings and civil structures. The building conditions are prone to environmental exposure and loadings factor that can cause cracks on the surface. To prevent the expansion of harm, effective treatment requires first the understanding of the cracking root cause, and then, a strategy for repair is implemented. Traditional approach of manual crack detection requires more effort and time. In this project, an automated vision-based crack detection system using deep learning framework is developed to assist the repair work in the term of resource planning. The encoder-decoder architecture in the deep learning framework produces convolutional features that improve the performance of image segmentation where the cracks can be differentiated from the background more effectively. View Project

Intelligent Classroom System

Artificial intelligent processing in video technology is growing rapidly, such as face identification and behaviour analysis. This project utilizes face identification features to process multiple camera views, update students’ attendance to a cloud database and implement Anomaly Activity Detection Module. It constantly tracks the students’ whereabouts and publish the updates of the students’ status to the same cloud database. In this project, ten test faces are detected, identified and recorded in database based on the trained dataset with high accuracy within an acceptable time frame across multiple camera views under certain conditions.
View Project 1 View Project 2

Deviant Scene Detection

Most of the organizations (e.g., educational institutions, manufacturing factories etc.) in Malaysia require supervisors or managers to monitor participants (e.g., workers) during the operating hours in order to ensure that participants comply with the rules and regulation. This project is expected to reduce the human effort in monitoring participants by replacing it with an intelligent camera that is able to detect abnormal activity. The way to implement this is to utilize the camera surveillance in the operation area by adopting technology such as artificial intelligence (AI), machine learning (ML), image & video processing or computer vision (CV) and Internet of Things (IoT) to learn and detect human activities/behaviours. View Project

Face Mask and Social Distancing Detection

Face masks have recently been a symbol of the global battle to prevent COVID-19 spread. The project is expected to develop an higher accuracy face recognition model to recognize human face with and without mask. Secondly, social distancing detector is to monitor if save distance is practised among people in public. Thirdly, face mask detection is to detect high-risk situations without proper mask. These technique can be used in various settings to assist decrease illness spread, such as classrooms, stores, and factories. View Project

Virtual Classroom Monitoring

The advancement of computer technology allows students to interact with educators using Artificial Intelligence (AI) technology through smart classrooms (E-Learning Classrooms). Currently, smart classrooms are believed to change the existing dull teaching methods and enhance the students’ learning experience. Therefore, the proposed system offers real-time users performance monitoring features such as detection and recognition on face and hand gesture to monitor student activities and recognize student behaviour through the smart classrooms. View Project

Zero-Shot Learning on Human Action and Gesture Detection

The Zero-shot Leaning module in an object detection process observes an input (e.g., video scene) is not being trained, and predicts its detected category. The team has applied this approach in a gesture-based human machine interaction (HCI) prototype solution. HCI studies continually emphasize the user experience especially when it is implemented in a real-world environment. As known that every individual acts differently and more uncontrollable environmental variables might affect the performance to detect and react to the gesture performed. Even though there are many solutions and datasets proposed in the market, not each of them perfectly fitted to our needs. Hence, to propose a more tailored made gesture detection for own use, the existing zero-shot learning model will be tested on the gesture dataset introduced in this work to fine tune to own needs. View Project

AR Furniture with 3D model captions

AR is one of the latest technologies that involves the integration of computer graphics with the user’s environment in real time. The proposed project is an AR mobile application which offers preview of furniture in one’s real environment by allowing the consumers to visualize how the particular furniture will look in the real world using a smartphone before they make a payment. To construct the furniture into a virtual object which is the 3D model file, photogrammetry is applied as it is able to perform 3D construction using a batch of images captured by the smartphone. When the construction is completed, the 3D model file will be converted into a manifest file and is ready to be uploaded to the cloud. In the system, whenever a user selects a model, the target model will be downloaded and markerless tracking will be activated to render the model in the user’s environment without a marker. View Project

AR Shoes with 3D model reconstruction

AR technology brings digital information and virtual objects into the physical space. With AR, the digital world comes to life inside the view captured by our tablet or phone camera. The objective of this project to be achieved is to enable the user to construct 3D shoe models without redesigning the 3D shoe models that they wanted to display in the AR application. Besides that, this project is used to try on the 3D shoe models virtually by using the AR technology. The purpose of this project is to let users try on the shoes more conveniently. They can just use their mobile phone to try on the shoes virtually anywhere and anytime. View Project

Visual Search Tool for E-commerce

This project aims to develop a mobile app for E-commerce with better customer experience. Hence, instance-based image retrieval is integrated into this app to help customers to search for similar products captured by them. This is especially useful for elderly people who may not be too tech savvy. It allows them to snap an image to find desired products instead of typing the product’s name. Image search is an option in the standard search box, allowing the user to snap a photo or select a photo from the camera roll to search. Then, it will sift through over items available in the product catalogue to find similar products. View Project

Care-U Application

Nowadays are exposed to both external and internal stress and more often than not, they deny its existence and take on destructive ways to deal with it. Counseling service in universities is generally not the first option nor the preference of the students as the satisfaction of realization of them now comes from social media. In this work, a mobile application inclusive of a Chatbot with sentiment analysis, blog and music sharing, and an interactive game was developed. This application serves as a platform for the students to share their thoughts freely where they would be comfortable being themselves. View Project

Cost Optimisation with Industrial Machine Performance

This project studies the production of a single type of product on multiple parallel machines by a manufacturer. The manufacturer has a pool of machines available with a combined production capacity greater than is required for production requirements. We assume that machine productivity may vary within a given range, which gives manufacturers the opportunity to adjust their total production capacity to meet demand and utilise the different cost structures of available machines. In this case, the manufacturer must decide which machine to choose to generate a given requirement and how to operate the machine. This project presents a deterministic mathematical model to support the production and distribution planning scenario. View Project

Image Processing Automation and PO Matching System

Nowadays, companies often process invoices, order forms, and other paperwork frequently. They are not satisfied with those common OCR technology to have only conversion of images into texts and look forward to having text recognition on required information from the documents and extract them out for validation purposes. The designed system is able to provide auto-validation for the documents such as Purchase Order (PO), invoice, and Delivery Order (DO). It aims to produce a customized OCR application for companies that could process different types of documents, retrieve the required information and perform auto validation. Besides that, this system is developed to provide functions such as data extraction and auto validation to various types of documents in order to reduce human involvement. The creating template features was also implemented to deal with any kinds of new documents in the future. By continuing to improve this project, the project is able to reach a satisfied accuracy level and be ready for real life implementation. View Project

High Efficiency Video Coding in Industry 4.0

The demand for multi-view (MV) has increased rapidly and there was a lot of research work to improve the technique and fulfil its needs. High Efficiency Video Coding (HEVC) compression standard has been implemented in this work. HEVC is a compression standard designed to reduce bitrate and remain the same quality compared to the previous compression standard Advanced Video Coding (H.264). It will provide a better compression to higher resolution video such as Ultra High Definition (UHD). In this paper present a preliminary study on MV with depth by using HEVC compression on a real-time streaming protocol. The study of proposed work may help the industry to enhance the viewing experience by multiple camera capture and also resolve the data traffic issue to transmit UHD video. MV & 3D-HEVC codec was also proposing to encode and decode in near real-time video streaming by setting up a three different view of cameras with depth prototyping on higher performance CPU. Few validation methods like BD-rate, time, QP and PSNR will be considered to be used to make a comparison with simulcast real-time multiple views architecture in difference HEVC extension. View Project