Automated Threat Detection in X-ray and CT Imagery for Advanced Border and Access Security

The continual increase in air travel has led to a growing concern regarding safety and security. To address this, security personnel at ports currently conduct manual inspections of incoming and outgoing passenger luggage, packages, and containers using x-ray scanners. However, this screening process is resource-intensive and demands constant vigilance from human experts. This approach introduces the risk of human error, stemming from factors such as fatigue, challenges in identifying contraband items, the need for rapid decision-making, and varying levels of experience among staff. This project aims to develop innovative computer vision and machine learning algorithms designed to automatically identify contraband items in scanned X-ray images of passenger baggage and mail packages at various ports, including those for air, sea, and land travel.

Automated Threat Detection
Computer Vision for Digital Pathology

Computer Vision for Digital Pathology

In Digital Pathology, the analysis of histopathological images is crucial for supporting pathologists in their diagnosis. It aids in pinpointing potential disease locations, assisting with interpretation, and ensuring the accuracy of diagnoses. However, such modalities present an intricate landscape of cell communities numbering in the hundreds of thousands. These communities exhibit a diverse range of textures and shapes, making the identification and classification of abnormal tissues, such as potentially cancerous ones, challenging, even for seasoned pathologists. This project focuses on developing intelligent analytical tools for profiling the tissue microenvironment in Whole Slide Images and seamlessly integrating them into comprehensive systems for pathology detection and monitoring.

Radiology Medical Image Analysis

This field of medical imaging focuses on the advanced analysis of X-ray, MRI, and CT images to detect and quantify various types of pathologies while enhancing the accuracy and efficiency of diagnostic processes. We devise algorithms and machine learning techniques to develop robust frameworks capable of identifying a diverse range of medical conditions across different imaging modalities. The objective is not only to improve the accuracy of pathology detection but also to provide quantitative assessments, aiding in the characterization and monitoring of diseases. This research holds significant promise for advancing the field of diagnostic radiology, ultimately contributing to more effective and personalized healthcare solutions.

Radiology Medical Image Analysis
Intruder detection and tracking in a uniform appearance crowd

Intruder Detection and Tracking in a Uniform Appearance Crowd

Despite the abundance of uniform crowds in many contexts, e.g.in the UAE and Gulf regions, and the challenges they exhibit, little or nothing was done to address the problem of tracking a person in a uniform crowd. Ground robots have been deployed in some tracking contexts, but most of the cases deal with a non-uniform crowd. In addition, a moving robot might cause inconvenience and risks to the crowd during the tracking. A more suitable approach is to deploy an Unmanned Aerial Vehicle (UAV). A UAV-based system for intruder/suspect detection and tracking is invasive, having a nearly zero risk of collision, and most importantly, has more flexibility in tracking when it comes to a vision-based system. This project aims to develop a vision-based drone-person tracking system that employs advanced and customized deep-learning techniques for detecting and tracking a target person in a uniform crowd.

Vision-Based Flare Analytics

In the oil and gas industry flaring is the process that consumes waste gases in a safe and reliable manner through combustion in an open flame. Flaring occurs in several processes like well testing and production operations. It is routinely used to dispose of flammable gases that are either unusable or uneconomical to recover. A flare system consists of a flare stack and pipes that feed gas to the stack. Flare size and brightness are related to the type and amount of gas or liquids in the flare stack. An efficiently burning flare does not produce visible smoke or soot. However, when incomplete combustion occurs, black smoke can be produced, and which is caused by wind, water, impurities in the fuel, or poor mixing with air. During such incomplete combustion, flaring can produce carbon monoxide, unburned hydrocarbons, and other volatile organic compounds. These can have negative effects on public health, and operators must often adhere to governmental regulations that prevent black smoke at oil and gas production facilities. The goal of this research to develop a flare monitoring and measurement system for detecting, predicting, and eventually mitigating any potential abnormalities in the flaring.

Vision-Based Flare Analytics
Computer vision for smart farming

Computer Vision for Smart Farming

Within the domain of smart farming, our research focuses on two crucial aspects: 1) the detection of fruit and vegetable maturity, and 2) the identification of plant diseases. Our efforts involve the development of advanced computer vision techniques capable of accurately determining the ideal ripeness stage by analyzing visual indicators like color, texture, and shape. These solutions are designed to streamline harvesting processes with precise timelines and reduce agricultural waste. Additionally, we employ machine learning and pattern recognition to elevate the early detection of plant diseases, facilitating timely interventions that mitigate the risk of extensive crop damage while minimizing the need for extensive pesticide use.

Speech Audio Analytics

Speaker identification from emotional and noisy speech is a challenging task that has been gaining attention in the past few years. This is due to the fact that emotions and noise can mask the speaker’s identity, making it difficult for existing algorithms to identify him/her accurately. The problem is even more complicated when it comes to public spaces, where multiple people are talking at once, and background noise can be overwhelming. In this research, we are addressing part of these challenges, notably those related to person identification, emotion recognition, and speech segregation.

Speech Audio Analytics
Scroll to Top