http://hdl.handle.net/123456789/21324 2026-10-17T05:40:57Z 2026-10-17T05:40:57Z Muhammad Haris, 01-136221-041 Kaleem Yousaf, 01-136221-037 http://hdl.handle.net/123456789/21657 2026-08-27T06:58:06Z 2025-01-01T00:00:00Z

AI-Powered Bidirectional Sign Language Translator Muhammad Haris, 01-136221-041; Kaleem Yousaf, 01-136221-037 The communication between deaf and hearing people still raises obstacles in different settings such as education, work, and even public places. This is mainly due to the shortage of sign language interpreters and the low level of sign language skills among the general public. As a result, these obstacles often lead to unequal participation, delayed information access, and exclusive communication. However, the continuous integration of AI and computer vision technologies presents a possibility of creating automated translation tools that could make communication easier and more accessible. This project, driven by this requirement, offers an AI-enhanced two-way sign language translator that features real-time Sign-to-Text and Text-to-Sign translation utilizing common hardware and a web-based interface. The Sign-to-Text component of the system makes use of a standard webcam to peg hand movements and sends each of the sequential frames through the MediaPipe Hands algorithm in order to detect the 21 main hand points. The identified points are then mapped into normalized feature vectors and subsequently categorized using a CNN (convolutional neural network) that has been trained on appropriate ASL alphabet datasets. Such a setup enables prompt detection of non-verbal random letter gestures without making use of particular sensors like data gloves or depth cameras. The result is gradually built up into a highly readable text, thus opening up the world of sign language to hearing users who can then follow the conversation of deaf users. The Text-to-Sign module acts as a reversal by issuing a transformation of written text into a series of ASL gesture images. This image-oriented approach gives a lightweight and resource-saving alternative to 3D animation, thus allowing it to work under low-power conditions while being clear to deaf users. The entire system is developed as a web application using the Django framework, which means it is accessible from any device and browser, therefore, no installation is needed. To test the performance of the system regarding recognition accuracy, latency, usability, and compatibility, extensive tests were performed. The Sign-to-Text classifier during controlled evaluations reached an accuracy of about 95%, with an average time of fewer than two seconds per prediction. Alongside this, the system was perceived as very easy to use by the deaf and the hearing participants; they all concluded that the interface is user-friendly, the translation results are easy to follow, and the system allows for basic communication to run smoothly. Compatibility tests also confirmed that the system performs reliably across the most popular web browsers and operating systems Supervised by Ms. Sabira Feroz

2025-01-01T00:00:00Z Muhammad Waleed Raza, 01-136221-023 Huzaifa Inam, 01-136221-009 http://hdl.handle.net/123456789/21655 2026-08-27T06:52:59Z 2025-01-01T00:00:00Z

Social Sentry: AI-Based Cyber Attack Detection of Phishing Links and Emails Muhammad Waleed Raza, 01-136221-023; Huzaifa Inam, 01-136221-009 Phishing is among the most widespread and harmful cyber threats, exploiting human trust to obtain credentials, financial data, and confidential information. This project introduces Social Sentry, an intelligent phishing detection system designed to identify malicious URLs and email content using modern deep learning techniques. The system incorporates two dedicated datasets: one consisting of phishing and legitimate URLs, and another containing phishing and benign email samples. These datasets enable Social Sentry to learn distinct patterns associated with both web-based and text-based phishing attacks. Over time, phishing has evolved into sophisticated campaigns utilizing automation, AI-generated text, cloned websites, and highly deceptive communication patterns. As traditional blacklist-based or rule-driven systems struggle to detect such modern threats, advanced machine-learning models capable of adaptive and context-aware analysis have become essential. The integrated platform includes URL analysis powered by a TabNet classifier and email phishing detection implemented through a Wide and Deep neural network, enabling the system to produce accurate predictions through feature-aware learning. A Python Django REST backend manages prediction requests, authentication, logging, and access to machine-learning models, while a responsive React frontend delivers intuitive real-time interaction. To promote safer user decision-making, Social Sentry incorporates precautionary mea sures that alert users when a prediction is classified as phishing. Instead of integrating any external explanation API, the system provides clear warnings, recommended safety actions, and contextual cues to help users understand potential risks associated with suspicious URLs or emails. A continuous feedback loop allows users to validate results and contribute new samples for model retraining. Experimental evaluation demonstrates strong performance across both datasets, with high AUC scores and reliable real-time inference. Overall, Social Sentry provides a scalable, interpretable, and user-friendly phishing detection solution suitable for educational, corporate, and individual cybersecurity en vironments. Its modular design supports future enhancements such as browser exten sions, multilingual detection, attachment analysis, and integration with additional threat intelligence sources. Supervised by Dr. Sadia Nazim

2025-01-01T00:00:00Z Ahmed Raza, 01-136212-056 Hamza Saleem, 01-136212-057 http://hdl.handle.net/123456789/21653 2026-08-27T06:46:11Z 2025-01-01T00:00:00Z

Deepshot Ahmed Raza, 01-136212-056; Hamza Saleem, 01-136212-057 The whole area of Football video analysis turned out to be very important for researchers because the necessity of speedy and automated ways of highlight generation is growing rapidly across digital media platforms. The longer the match, the more manual review becomes time-consuming, inconsistent, and therefore unsuitable for large-scale workflows. The project called "DEEPSHOT" presents an intelligent, fully-automated solution that is able to transform raw soccer match recordings into concise highlight reels by integrat ing three main technologies: deep learning, temporal modeling, and multi-stage video analytics. Different from regular systems which heavily depend on the creation of rules and the use of simple motion detection, DEEPSHOT is a machine learning method that is based on deep visual embeddings and the interpretation of temporal sequence. Extraction of frame-level representations is the first step of the system, which is achieved by using a pretrained ResNet-152 backbone, followed by dimension reduction to create compact temporal descriptors for processing long-sequence. After that, the descriptors are processed by a multi-stage temporal modeling pipeline to detect key soccer events like goals, shots, fouls, cards, corners, and substitutions with a very high level of confidence. The system integrates multiple processing units which include: feature extraction, PCA based compression, temporal event scoring, time stamp alignment and automatic highlight assembly to comprise the one workflow. DEEPSHOT not only gives structured JSON event metadata but also uses FFmpeg video automation to create a polished highlight reel without any human participation. The whole system is developed using Python and is also supported by a mixture of well-known libraries like PyTorch, NumPy, SciPy, scikit-learn, OpenCV, which guarantees it to be efficient, scalable, and production-ready. Through the application of deep visual perception and advanced temporal analysis, DEEPSHOT has developed a very powerful and flexible solution for soccer highlight generation. It indeed lessens the manual editing and at the same time maintains the same quality of highlights and demonstrates the real-world applicability of AI video summarization in modern sports media technology Supervised by Mr. Abdul Salam

2025-01-01T00:00:00Z Muhammad Nauman, 01-136221-016 Ali Abbas Hussnain, 01-136221-032 http://hdl.handle.net/123456789/21650 2026-08-27T06:38:03Z 2025-01-01T00:00:00Z

Intelligent Automated Attendance System Using Real Time Facial Recognition Muhammad Nauman, 01-136221-016; Ali Abbas Hussnain, 01-136221-032 Bahria University, until now, has used primitive methods for attendance, such as; roll calls and paper signatures. Attendance tracking might sound like a simple and mundane task, but it is plagued by many errors which are surprisingly tricky to get right with the traditional methods being used. The idea of an Intelligent Object Tracking System with a Pan-Tilt camera for the University is presented in this thesis. Instead of physical sign sheets or simple biometric scanners, the projected system allows real-time facial recognition that is non-intrusive and unbiased to external conflicts. The system combines MTCCN face detection with ResNet embeddings with two separate user interfaces, making it a dual-interface architecture. A supreme administrative control is provided by a Streamlit dashboard while monitoring is influenced with the help of a Flask server. Live video feeds can be tracked and monitored by the instructors over local networks without the need of special equipment or the need to run back to administrative work stations. A PostgreSQL database makes it easy to manage the attendance records and support any needed analysis. The system hit a 97.2% of accuracy during its trial testing. Supervised by Ms. Nabia Khalid

2025-01-01T00:00:00Z