Artificial Intelligence Department

Our innovative model revolutionizes autonomous vehicle perception by unifying multiple critical tasks—traffic object detection, drivable area segmentation, and lane line detection—into a single, lightweight deep learning architecture. Designed specifically for real-time, embedded deployment on resource-constrained platforms like Raspberry Pi 5, YOLOPX delivers high accuracy and efficiency without the latency and redundancy common in traditional multi-model systems. This multi-task approach enhances safe navigation and situational awareness in complex urban environments, making it ideal for next-generation autonomous driving solutions. Designed for autonomous vehicle developers, embedded system integrators, and smart mobility platforms, YOLOPX pushes the boundaries of edge AI for safer, smarter transportation.

This project develops a state-of-the-art deepfake detection system that combines the fast and precise face localization capabilities of a modified YOLO architecture with advanced spatial-temporal feature extraction techniques to identify manipulated facial content in images and videos in real time. By integrating convolutional neural networks for detailed spatial analysis with recurrent neural networks for capturing temporal inconsistencies across frames, the system robustly detects subtle deepfake artifacts indicative of forgery.

Designed for rapid, high-accuracy detection to combat misinformation and digital fraud, this YOLO-based DeepFakeGuarddetector balances speed, accuracy, and scalability, making it a powerful tool against evolving synthetic media threats.

Our AI-Based Child Detection and Alarm System for Pool Safety leverages advanced computer vision with a YOLOv5 object detection model and a custom neural network classifier to accurately identify children near swimming pools in real time. Running efficiently on a Raspberry Pi 5, the system continuously monitors predefined danger zones around the pool and immediately triggers audible and visual alarms when a child enters these high-risk areas, providing a proactive safety layer beyond traditional fences and supervision. Designed for low latency and high accuracy with affordable hardware, this intelligent solution enhances drowning prevention by offering continuous, autonomous monitoring and instant alerts to caregivers, significantly reducing the risk of accidental pool drownings in residential and public settings.

AirDrawing is an advanced gesture-based drawing system that transforms natural hand movements into real-time digital sketches in midair, using only a standard webcam and computer vision. By integrating hand landmark detection with gesture recognition, AirDrawing enables users to draw, select tools, and erase without physical contact or specialized hardware. The system overlays drawings onto live video feeds, enhancing interactivity in remote communication and e-learning environments. AirDrawing incorporates robust alphanumeric character recognition through specialized convolutional neural networks, enabling accurate interpretation of digits and letters drawn in 3D space. This innovation expands AirDrawing from a simple sketching tool into an intelligent interface with applications in video conferencing, education, and metaverse text input, offering an accessible, low-latency solution for intuitive human-computer interaction. In fact, it pushes the boundaries of touchless interaction by combining real-time hand tracking, gesture control, and machine learning into a seamless, user-friendly platform.