Pedestrian tracking, as a sub-task of object tracking, is widely applied in smart surveillance. However, the accuracy of object re-identification significantly decreases when objects encounter prolonged occlusion or undergo significant appearance changes. To address this issue, PTNet, a regression-based object tracking framework, is proposed in this paper. Firstly, a combination of DeepSORT and FastReid models is employed to resolve the problem of target re-identification after being occluded for a period of time. Secondly, a self-attention-based feature extraction model named BoTNet, incorporating the Global Multi-Head Self-Attention (MHSA), is integrated into the backbone network to capture global and long-distance features of targets in surveillance videos. Through ablation experiments, it was found that PTNet enhances accuracy by 4.8% compared to the benchmark. When compared to state-of-the-art object detection algorithms, PTNet demonstrates a significant improvement in performance.
The Abstract Alzheimer's Dementia (AD) presents significant diagnostic challenges, particularly in terms of early detection, where traditional methods often fall short due to their invasiveness and high costs. This study introduces a novel, non-invasive approach utilising emotional expressions captured from audio recordings to detect AD. Employing advanced digital signal processing techniques, including Facebook's Denoiser model, and deep learning methodologies through models such as Wav2Vec 2.0, this research aims to identify emotional disturbances that precede cognitive decline. Audio recordings were transformed into a tabular format, suitable for machine learning analysis. The LGBM Classifier and ensemble methods demonstrated superior performance, with the LGBM Classifier achieving the highest F1 score of 0.93 and an accuracy of 0.89 on a 3.5-second segment. These findings underscore the potential of combining emotional analysis with machine learning to enhance early AD detection, offering a simpler, more accessible diagnostic tool than currently available methods.
This study explores the potential of acoustic features extracted from speech recordings for detecting Alzheimer's Dementia (AD), employing a comprehensive approach that incorporates binary classification (healthy control vs. dementia), multiclass classification (healthy control, mild cognitive impairment, AD), and regression analyses (predicting MMSE scores). Additionally, demographic information of the participants was integrated to enhance the models' predictive accuracy. Our methodology involved processing each dataset version through a series of machine learning models tailored to each task, starting with a baseline version, followed by hyperparameter optimisation, and finally applying a combination of preprocessing steps (scaling, outlier removal, dimensionality reduction, and skewness correction) to identify the optimal setup for each model. The findings indicate that preprocessing steps significantly improve model performance across all tasks, underscoring the importance of data preparation in machine learning workflows for healthcare applications. Notably, the use of acoustic data alone for AD detection shows promising results, suggesting a pathway toward more generalised approaches that could incorporate recordings in various languages without linguistic dependency. This opens up the possibility for scalable, non-invasive screening tools for AD, leveraging the universal nature of acoustic markers in speech for early detection and monitoring of this condition.
Learning from Demonstration (LfD) techniques are invaluable for capturing complex human behaviors for robotic arm manipulations, yet they frequently encounter challenges such as avoiding singularities and respecting joint limits when directly applied to robotic systems. These challenges often lead to mechanical non-compliance, inaccuracies, and increased computational demands during control method adjustments. To address these issues, this study introduces a robust approach that integrates Damped Least Squares Inverse Kinematics (DLS-IK) with Probabilistic Movement Primitives (ProMPs). By leveraging DLS-IK to generate kinematically feasible trajectories, and embedding these within the ProMPs framework, our method not only ensures mechanical compliance but also capitalizes on the probabilistic modeling capabilities of ProMPs. This synergy addresses a significant gap in traditional LfD applications—aligning human demonstrations with the mechanical constraints of robotics, independent of the demonstrator’s expertise. Our integrated approach refines the LfD process, enabling the generation of precise, reliable, and mechanically compliant movements in robotic arms, thereby reducing the typical inaccuracies and computational burdens associated with conventional LfD methods.
This paper reviews the impacts of user behavior, dwelling characteristics, and social networks on energy savings within the UK housing sector. The study highlights the role of strategic energy planning, emphasising stakeholder engagement and the potential of social networks to influence energy-saving practices. It shows that integrating behavioral insights with architectural design is beneficial to enhancing energy use efficiency and reducing energy cost. By examining the interplay between human behavior and technology, a holistic approach to energy conservation is advocated integrating technological innovations and social impacts that drive sustainable energy usage.
