Signal Processing

Signal and Information Processing is particularly well suited to deal with multimodal data. The current and future sensor systems e.g., sensors in smart devices, will provide ever more data for subsequent analysis; hence, the research motivation is multimodal (audio-video-infrared-touch-text) data processing for next generation of artificially intelligent automated systems.

We aim to contribute AI that will enable significant advances towards the vision. The specific aims for Healthcare and Security and Surveillance are:

1. Intelligent Sensing Led AI for Mental Health
2. Multimodal Human Behaviour Analysis Including Anomaly Detection
3. Multiple Human Identification and Tracking in Cluttered and Congested Environments
4. Multimodal Speech Enhancement and Separation in Challenging Environments

YouTube Video: Signal and Information Processing

Intelligent Sensing Led AI for Mental Health

Intelligent Sensing (audio-video-touch/keypad-text i.e., multimodal sensors) to capture attention deficit hyperactivity disorder (ADHD) health symptoms.

Trustworthy AI (machine learning/deep learning) to help clinicians to perform the ADHD early diagnosis and thereby administer effective treatment.

Collaboration with Cumbria, Northumberland, Tyne and Wear NHS Foundation Trust.

Multimodal Human Behaviour Analysis

Human anomaly detection is a challenging task for reliable artificial intelligence (AI) because the behaviour of individuals and group of humans requires robust features for modelling, indexing, and classification. Therefore, this research is on multimodal action recognition, contextual information retrieval and information fusion for human behavior analysis. 

Advanced deep learning techniques are researched with the main focus to provide network explainability and human informed system.

Please see our other YouTube videos on some of our other research topics below:

• Multi-feature Fusion for Abnormal Event Detection
• AnomalyXPose
• ActionXPose

 

Multimodal Human Identification and Tracking

Research involves core signal processing for multiple human localization and tracking in cluttered and congested environments.  

The core signal processing techniques providing the solution for MHT (multiple human tracking) include:

• social interaction modelling
• probabilistic clustering
• non-linear filtering
• efficient data association
• stochastic differential equations based particle flow
• multi-level cooperative fusion

Please also see our video on Online Multiple Human Tracking for more information on this area of research.

 

 

Multimodal Speech Processing

We have made major breakthroughs in establishing complete mathematical proofs for nonlinear signal processing theory. This has enabled us to establish non-linear models for optimal signal separation and information retrieval.

One of our successes is the development of a complete framework using statistical methodologies for signal fusion, separation of complex nonlinearly mixed signals and informational retrieval. This work has radically challenged conventional approaches. Research for speech and audio show outstanding performance for overdetermined, determined, underdetermined (including binaural and single channel) cases.   

The work has contributed to the fundamentals of nonlinear signal processing theory and exploited advanced machine learning techniques to pave the way for the research community to interdisciplinary solutions.