Department Projects
Project Title | GPS-Aided Visual-Inertial Odometry for Self-Driving Car |
---|---|
Group Members |
|
Project Supervisor(s) | Dr Fahad Mumtaz Malik |
Project Abstract | The project “GPS-Aided Visual-Inertial Odometry for Self-Driving Car” focuses on enhancing the navigation and localization system for self-driving cars to improve their precision and operational efficiency. The project addresses a key challenge, which is the alleviation of drift in state estimation, achieved through the implementation of a GPS-aided visual-inertial odometry technique. By fusing GPS sensors with the Visual Inertial Navigation System (VINS), the state estimation of self-driving vehicles is significantly enhanced. This is accomplished using strategically placed sensors and a GVINS model, which computes the vehicle’s final state estimates. An Extended Kalman Filter (EKF) is employed to refine predictions using data from GPS and VINS, significantly improving accuracy and helping address key challenges faced by autonomous vehicles. |
Project Title | Specific Emitter Identification |
---|---|
Group Members |
|
Project Supervisor(s) | Dr Qasim Umar Khan |
Project Abstract | Specific Emitter Identification (SEI) refers to the identification of transmitters based on their unique characteristics, known as RF fingerprints. SEI can be performed using two main methods: manual feature-based and deep learning-based approaches. In this project, we are developing a novel SEI algorithm that uniquely combines both approaches.
First, the signal is acquired using a software-defined radio. The received signal is then passed through the SEI algorithm, which extracts RF fingerprints from it. The signal is decomposed into five modes using Variational Mode Decomposition (VMD). From each mode, two types of features are extracted: RF-DNA and time-frequency spectrograms. We refer to this combination of RF-DNA with VMD as modified RF-DNA, and it serves as the input to the XGBoost classifier, which classifies the signal among known classes. If an unknown transmitter is detected, it is sent to the Siamese Neural Network (SNN). The SNN uses both modified RF-DNA and time-frequency spectrograms to convert the inputs into a shared representation space using a 2-channel Convolutional Neural Network (CNN) with bimodal feature fusion. The similarity scores are then calculated by comparing the input with other signals. If a match is found, the transmitter is labeled accordingly; otherwise, it is assigned a new label as an unknown transmitter. |
Project Title | Breast cancer detection from mammograms using deep learning. |
---|---|
Group Members |
|
Project Supervisor(s) | AP Sobia Hayee |
Project Abstract | Breast cancer detection is a critical task in medical imaging, with mammography being the primary screening tool. In recent years, deep learning techniques have shown promise in improving the accuracy and efficiency of breast cancer detection from mammograms. This study proposes a deep learning model for breast cancer detection from mammograms, leveraging convolutional neural networks (CNNs) for feature extraction and classification. The model is trained on a large dataset of mammograms annotated by expert radiologists. Experimental results demonstrate the effectiveness of the proposed deep learning model in accurately identifying suspicious regions indicative of breast cancer, achieving competitive performance compared to traditional methods. The proposed approach holds significant potential for enhancing breast cancer screening programs, aiding radiologists in early detection and diagnosis, and ultimately improving patient outcomes. |
Project Title | Smart Cane for Visually Impaired |
---|---|
Group Members |
|
Project Supervisor(s) | AP Sobia Hayee |
Project Abstract | This project presents a smart cane designed to enhance the mobility and safety of individuals with visual impairments. The cane integrates a suite of sensors, notably ultrasonic sensors and a camera, to detect obstacles. Utilizing a Raspberry Pi 5 as its central processing unit, the cane employs a convolutional neural network for object recognition through its camera.
Upon detecting obstacles within a range of 3 feet, the ultrasonic sensors trigger the activation of the camera, initiating a feedback loop. This feedback is relayed to the user through multiple modalities: clear and concise audio instructions, emitted via an earpiece connected through Bluetooth; audible alerts from a buzzer, offering immediate attention in noisy environments; and tactile cues via haptic feedback delivered through strategically placed vibrators on the cane. The cane is 3D printed to assemble it into a fully functional hardware form, ensuring precision and durability in its design. By combining advanced sensor technology with multi-sensory feedback mechanisms, this smart cane represents a significant advancement in assistive technology, offering visually impaired individuals a comprehensive solution for navigating their surroundings with confidence and ease |
Project Title | Fixed Wing Variable Sweep UAV |
---|---|
Group Members |
|
Project Supervisor(s) | Dr Fahad Mumtaz Malik |
Project Abstract | This project aims at the development, analysis, and practical deployment of a variable sweep fixed-wing UAV, with a focus on improving adaptability and performance across swept wing conditions. The main challenge is to stabilize the UAV using a custom-built flight controller in maintaining positive static stability (level flight) with swept and unswept wing configuration, using control theory. A sophisticated control system is developed to manage stable and straight flight during wing-sweep. Such a UAV has implications for both civilian and defense sectors such as corporate transportation/delivery, agriculture, surveillance and reconnaissance. This project underscores the transformative potential of variable sweep fixed-wing UAVs in the realm of unmanned aerial systems for
optimising flight parameters autonomously. |
Project Title | Solar-Wind Hybrid Power System with Battery Storage |
---|---|
Group Members |
|
Project Supervisor(s) | Dr. Taosif Iqbal |
Project Abstract | This project aims to design a multi-port DC-DC converter (DC micro-grid) that integrates renewable energy sources such as solar and wind, along with a battery as energy storage element. The goal is to create a robust system that maintains constant output voltage across the DC bus. The algorithm involves a peak current mode controller implemented using Arduino. |
Project Title | Autonomous Floor Cleaning Robot |
---|---|
Group Members |
|
Project Supervisor(s) | AP Sobia Hayee |
Project Abstract |
Traditional cleaning processes, while effective to some extent, often fall short in addressing specific needs such as allergic cleaning and daily maintenance. They tend to be time-consuming and labor-intensive, posing challenges for individuals with busy schedules. Our project focuses on advancing vacuum cleaner technologies, exemplified by the Roomba, through integration of Embedded systems, Computer vision, AI, Visual SLAM Implementation, Power Electronics, and Control systems. Utilizing the Raspberry Pi 4B+ platform with a webcam, sonar sensor, and IMU. visual SLAM, a cost-effective alternative to Lidar SLAM, enables precise localization and mapping, enhancing adaptability and effectiveness. Our robot targets small spaces, serving pet owners, allergy sufferers, and venues. We tackled challenges like power optimization and sensor fusion, starting from market surveys to final model printing in AUTOCAD, providing a precise and affordable cleaning solution. |
Project Title | Fleet Management System |
---|---|
Group Members | Shayan Ali Sadiq
Amaad Ali GC Haider Imtiaz GC Haris Ahmed |
Project Supervisor(s) | Dr Qasim Umer Khan |
Project Abstract | Fleet Management system Commercial engine vehicles, for example vehicles, vans, trucks, expert vehicles, forklifts, and trailers Private vehicles utilized for work purposes Aviation apparatus, airplane Ships Rail vehicles.
With the ascent in the number of business Fleet vehicles consistently, it is hard for the Fleet supervisors to utilize conventional and manual methods of dealing with its Fleet. Utilizing the present mechanical progressions, it is conceivable to grow such a framework, that deals with all the concerns of the Fleet chief. Developed innovations like Global Positioning System (GPS), Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), and WI-FI can be used to devise such a framework. Knowing where the vehicles are, what the drivers are doing, and checking each occasion progressively is the key boundary for an all-around oversaw dynamic procedure. The proposed Intelligent Fleet Management System will permit Fleet administrators to distantly follow vehicles utilizing reasonable guides and perspectives produced from Global Positioning System (GPS) information. Also, it will likewise be equipped for giving ongoing rate, continuous forecast of appearance time at the goal, driving conduct of the driver and convenient support alarms, and so on. These additional highlights will make this framework insightful and shrewd. This framework will help in getting more authority over drivers and vehicles, forestalling delays in conveyances, improving driving propensities, forestalling questionable revealing and decreasing upkeep costs, and so on. |
Project Title | Power Monitoring and Optimization System |
---|---|
Group Members | Adil Akhtar Ali khan
Syed Haannie Ali Kazmi Muhammad Abdullah Naeem |
Project Supervisor(s) | Dr Usman Ali |
Project Abstract | With the usage of electrical appliances and various electrical devices, there is a constant need of knowing where and why the power consumption is excessive. This project aims to monitor the electrical power consumed by different electrical devices; whether at home, office or in an industry. Then notify the user about the how much voltage is being provided by the electric supply companies; how much current is being consumed by different appliances and how much the overall power is being consumed. The project uses a standard Arduino Uno to collect data from its voltage and currents sensors. The micro controller processes the information and provides the results on android application, LCD (Liquid Crystal Display) and on a Serial Monitor on a Computer. RMS (Root Mean Square) voltages and currents are measured to provide the best readings in RMS values. The second part of the project includes the optimization phase. It is to optimize the loads if the power consumption increases from a certain current value. Relays shall operate to counter that. The system is automated. |
Project Title | Car Make and Model Recognition using Deep Learning |
---|---|
Group Members | Amna Saleem
Abdullah Saad Syed Muhammad Usman Ghani |
Project Supervisor(s) | Asst. Prof. Sobia Hayee |
Project Abstract | Security and Surveillance is one of the major problem. Due to human error the degree of security is at risk. Car make and model recognition system will assist the security agencies to catch the culprit as soon as possible by identifying the make and model of the car. It can help many businessmen who wants to open car show room as they can observe the new trends through Car Make and Model Recognition System. Though descriptors such as SIFT, SURF, HOG are used to extract features or key points of the image but it cannot be applied on thousands of images due to computational problems. A system with heavy RAM and multiple GPUs will be required so ‘Deep Neural Network’ are very helpful in this case. We trained Stanford car dataset on ResNet50 model. Performed training with and without augmentation to observe the difference in the accuracy of the model and the classifier. Classifier accuracy with data augmentation is 87.60% and without data augmentation is 77.91%. |
Project Title | Detection of Abnormalities in Chest Radiographs Using Deep Learning |
---|---|
Group Members | Aiman Tariq
Azka Rehman Mushkbar Fatima |
Project Supervisor(s) | Asst. Prof. Sobia Hayee |
Project Abstract | A chest x-ray radiograph is a crucial tool to diagnose numerous diseases. Diagnosing and treating patients timely is crucial which results in overworking the radiologists. With the development of AI and particularly deep neural networks, it has become possible to assist different professionals with their tasks. This project aimed at training convolutional neural network on the dataset containing thousands of chest x-ray images for finding chest abnormalities. DenseNet121 was selected for training. Chexpert, a large chest radiograph dataset from Stanford was acquired for training. The trained model on this dataset optimizes the loss function and reports the test accuracy of 81% with dataset having unbalanced classes but with poor f1 score. So, the model was trained again on balanced classes and f1 score was significantly improved with test accuracy of 90%. |
Project Title | Micro Inverter with Lithium Ion Battery |
---|---|
Group Members | Muhammad Danial Khilji
Furqan Ahmed Chishti Fidaullah Noonari |
Project Supervisor(s) | Dr Usman Ali
Dr Aqib Pervwaiz |
Project Abstract | There is an immense need of a small size of inverter globally. Nowadays an inverter is an important element for use as a back-up power supply and it is also an essential part in generating off-grid power through solar panels. Inverters available in the market are considerably big which takes up a lot of space. Most of the commercial inverters available in Pakistan are simply PWM inverters. These inverters have high harmonics and are bulky to move around. Furthermore, they use car batteries as their battery which are dangerous to place in open home environment. These inverters are not only enormous in size but also inefficient for the office equipment use. Most importantly they don’t have an ability to control abrupt power surges.Our main objective was to successfully reduce the size of the inverter while maintaining the efficiency of 220V and 150W which is intentionally designed to work for IT equipment. An inverter is used to convert DC voltage from the battery to AC voltage of the main supply. It also charges up the battery when not in use. Our inverter also filters out the abrupt power surges from the main supply power. This inverter was specifically designed as the universal supply. Its main use will be as an office equipment back-up supply, therefore requires fine sine wave at all time. The technique used is push-pull for the buck and boost as well. We used the Lithium Ion batteries instead of conventional chemical batteries which takes a lot of room space. Lithium Ion batteries are not only small in size, they also have a greater life cycle with very low or no maintenance at all. |
Project Title | Hybrid Solar Inverter |
---|---|
Group Members | Abdullah Shehzad
Hamza Imran Khalid Hussain |
Project Supervisor(s) | Dr Usman Ali |
Project Abstract | Our objective was to propose a multi-input DC-AC inverter for hybrid PV and Grid powered system which consists of a multi-input DC-DC fly-back converter, a three phase full-bridge DC-AC inverter and a Variable-frequency drive system in order to produce a constant output voltage from the different energy sources which was being used to control the speed of an induction motor. The initial application and aim of the project was to use this system to control the speed of the compressor motor of an Air-conditioning unit, however it can be modified for various other applications. The entire model was simulated on the MATLAB software, which showed successful results for the implementation of the system. |