Global road traffic accidents remain a significant problem resulting in large numbers of fatalities and economic costs. Automatic accident detection from traffic surveillance cameras can enable rapid emergency response to save lives and reduce traffic disruptions. In this paper we present our new deep learning architecture for the detection of road accidents from CCTV footage, called Enhanced Spatial Accident Detection Network (ESADN). In our approach we introduce three innovative concepts: a Transformer-inspired Spatial Attention Mechanism for modelling contextual relationships between image regions, a Feature Enhancement Module (FEM) that enhances the spatial features of accidents, and a Context Aware Classification Head for improving classification accuracy through feature fusion. The superiority of ESADN is demonstrated by extensive experiments on the Road Accidents from CCTV Footages Dataset (6,191 accident vs. 15,420 non-accident images): it achieves 99 % accuracy, 0.98 F1-score on the accident class, ROC-AUC = PR-AUC = 1.00, and an average precision of 1.00. When exported to ONNX and run with ONNX Runtime, the model runs at ∼6.6 FPS (0.152 s/frame) and has a 105 MB footprint. Visualization of activation maps confirms that ESADN consistently focuses on collision zones and anomalous vehicle positions. ESADN’s high accuracy and edge-readiness make it an ideal candidate for real-world traffic monitoring systems.

The complexity of predicting wine quality arises from the relationships between physicochemical variables, many of which are intercorrelated and redundant. This paper proposes an integrated wine-quality prediction framework that applies feature compression to obtain a compact representation of the original feature set, and combines this representation with the wine volume information. The compressed feature set is then utilized to train a set of individual classifiers, including Random Forest, Gradient Boosting, XGBoost, and CatBoost. This study combines feature compression with a stacking ensemble approach, which integrates the outputs of individual classifiers and determines the final prediction using Logistic Regression as the meta-classifier. The proposed method consistently achieves superior prediction accuracy and stability compared to individual classifiers by employing feature compression within a stacking framework. Ultimately, this work offers an effective, scalable, and interpretable solution for wine quality evaluation in both practical applications and decision-support systems.

Brain tumors remain a major neurological challenge, where timely and accurate diagnosis is critical for improving patient outcomes. Although several reviews have examined machine learning (ML) and deep learning (DL) techniques for brain tumor analysis, most existing surveys either focus on a single methodological family or lack a comparative perspective across emerging computational paradigms. This review addresses that gap by providing an integrated analysis of ML, Convolutional Neural Networks (CNNs), Transformer-based models, Generative Adversarial Networks (GANs), and hybrid ensemble frameworks for tumor detection, classification, and segmentation using magnetic resonance imaging (MRI). Unlike prior reviews, we systematically evaluate the clinical applicability, dataset limitations, and reproducibility concerns of these models while identifying unresolved issues such as interpretability, data scarcity, and domain generalization. Furthermore, we synthesize trends in multimodal learning, federated frameworks, and explainable AI, offering actionable insights for translating research advances into clinical practice. This critical perspective highlights not only the state of the art but also the pathways required for developing robust, transparent, and clinically viable artificial intelligence (AI)-driven diagnostic systems.

Accidental falls among the elderly present serious health risks and are a significant concern, particularly for individuals living alone. Annually, approximately 2.8 million seniors require emergency medical attention due to fall-related injuries, highlighting the urgent need for effective fall detection and response mechanisms. While video-based fall detection systems tend to be more expensive than wearable solutions, their ability to integrate with smart home technologies enhances their practicality and real-time monitoring capabilities. This review systematically examines video-based fall detection methodologies, assessing their effectiveness, challenges, and constraints across different processing stages. Furthermore, we provide a comparative analysis of state-of-the-art techniques, identifying key advancements and potential areas for future research to improve reliability and accuracy.

Accidental falls and their consequences are critical concerns for elderly people. Fatal injuries, when delayed in treatment, can lead to severe outcomes. Fall detection systems are crucial for the timely treatment of such injuries. Although sensor-based fall detection approaches are effective, video-based approaches are more useful because they assist in analyzing the fall scene and identifying the cause of the fall. However, privacy preservation is a major concern in video-based fall detection. The proposed system introduces a privacy-preserving mechanism that masks the identified human with a silhouette. A custom dataset, including 80 activities of daily living and 70 fall activities, is introduced. An LSTM variational autoencoder architecture is designed with a gradient clipping mechanism and a smooth variant of Adaptive Moment Estimation with Stochastic Gradient Descent (AMSGrad) optimizer to enhance the accuracy of fall detection. The reconstruction error between normal and fall activities is clearly identified with the help of a dynamic threshold. This results in a system performance that achieves accuracy, precision, and sensitivity of 99%, 97%, and 99%, respectively.

This study investigates the use of advanced machine learning (ML) algorithms to enhance classification accuracy in hyperspectral image analysis, addressing complexities across various datasets. Beginning with an exploration of hyperspectral imaging technology and its applications, the study preprocesses datasets to extract crucial features. State-of-the-art ML algorithms are then employed for training and evaluating classification models, allowing for comparative analysis of their strengths and weaknesses. Performance metrics, and confusion matrix are utilised to quantitatively assess algorithmic efficacy. The findings not only inform specific datasets but also highlight broader implications for hyperspectral image analysis. This research underscores ML algorithms' potential in extracting meaningful information from hyperspectral datasets, advancing understanding and offering practical implications for real-world applications, thus contributing to the field's ongoing advancement and enabling informed decision-making in diverse domains.

According to the World Health Organisation, falling is a major health problem with potentially fatal implications. Each year, thousands of people die as a result of falls, with seniors making up 80% of these fatalities. The automatic detection of falls may reduce the severity of the consequences. Our study focuses on developing a vision-based fall detection system. Our work proposes a new feature descriptor that results in a new fall detection framework. The body geometry of the subject is analyzed and patterns that help to distinguish falls from non-fall activities are identified in our proposed method. An AlphaPose network is employed to identify 17 keypoints on the human skeleton. Thirteen keypoints are used in our study, and we compute two additional keypoints. These 15 keypoints are divided into five segments, each of which consists of a group of three non-collinear points. These five segments represent the left hand, right hand, left leg, right leg and craniocaudal section. A novel feature descriptor is generated by extracting the distances from the segmented parts, angles within the segmented parts and the angle of inclination for every segmented part. As a result, we may extract three features from each segment, giving us 15 features per frame that preserve spatial information. To capture temporal dynamics, the extracted spatial features are arranged in the temporal sequence. As a result, the feature descriptor in the proposed approach preserves the spatio-temporal dynamics. Thus, a feature descriptor of size (Formula presented.) is formed where m is the number of frames. To recognize fall patterns, machine learning approaches such as decision trees, random forests, and gradient boost are applied to the feature descriptor. Our system was evaluated on the UPfall dataset, which is a benchmark dataset. It has shown very good performance compared to the state-of-the-art approaches.

Human activity recognition plays a prominent role in applications like sports, violence detection, accident detection, women security, and smart homes by predicting abnormal human behaviour. Human activity recognition now offers a solution for fall detection systems that assist older people at home. A fall detection system that responds promptly to fatal falls can help to reduce the fall severity. Fall detection systems are developed using wearable environmental sensors and video data. Wearable devices are not always feasible as they cause inconvenience to the user. Hence, our paper presents a new promising solution for fall detection using vision-based approaches. In this approach, we analyse the human joint points which are the prime motion indicators. A set of key points of the subject are acquired by applying the AlphaPose pre-trained network. These keypoints are inferred to be the joint points of the subject. The acquired keypoints are processed through a framework of convolutional neural network (CNN) layers. Here, the spatial correlation of the keypoints is analysed. The long-term dependencies are then preserved with the help of long short-term memory (LSTM) architecture. Our system detects five types of falls and six types of daily living activities. We used the UP-FALL detection dataset for validating our fall detection system and achieved commendable results when compared to the state-of-the-art approaches. For comparison, we employed the OpenPose network for keypoint detection. It is inferred from the results that AlphaPose network is more precise in keypoint detection.

Recognizing human activities from video sequences or sensor data is a challenging task in computer vision. Background clutter, partial occlusion, changes in viewpoint, lighting, and appearance are creating bottlenecks in the recognition of activity. In this paper, we provide a comprehensive review by categorizing the activity recognition approaches that have been applied on multivariate time series data. The review provides insights of each method, research issues and performance issue.

Security is one of the major components of the Smart Cities Mission of the Indian Government. This paper presents an efficient way of improving the existing security system and making the threat prevention and protection process more automated. We introduce a dynamic smart security camera system which has a trained image classifier model fed into it that observes and recognizes the movements and gestures of the human or object in its video range and detects any illegal or suspicious activity.