Computer Vision Techniques Research Articles

Enhancing Pose Estimation with Depth Integration: A Comparative Study of ArUco Markers and Depth Data

Pose estimation is a fundamental technique in computer vision, allowing systems to recognize and analyze the positions and movements of objects or individuals in three-dimensional space. Its applications extend to healthcare for physical therapy, sports for performance analysis, and surveillance for behavior monitoring. The development of advanced algorithms and hardware has significantly improved its accuracy and real-time processing capabilities, driving innovation across industries. This study examines the integration of depth data with ArUco marker detection to improve pose estimation precision. By comparing translation vector (tvec) calculations using RGB-only methods and depth-enhanced approaches, this paper demonstrates that depth data provides superior consistency, particularly at extended distances. The results suggest that depth integration could address the limitations of traditional marker systems, improving accuracy and robustness. This work highlights the potential of depth-based enhancements and outlines future directions for comprehensive evaluation and system refinement.

Linguistic to 3d Transformation

Converting linguistic input into 3D models is a complex and multifaceted task that requires the integration of several advanced technologies. Natural language processing (NLP) is used to understand and interpret the user's descriptions, while computer vision techniques help in recognizing and processing visual information. 3D rendering techniques are then employed to create the final 3D models.

Individual Cow Recognition Based on Ultra-Wideband and Computer Vision

This study’s primary goal is to use computer vision and ultra-wideband (UWB) localisation techniques to automatically mark numerals in cow photos. In order to accomplish this, we created a UWB-based cow localisation system that involves installing tags on cow heads and placing several base stations throughout the farm. The system can determine the distance between each base station and the cow using wireless communication technology, which allows it to determine the cow’s current location coordinates. The study employed a neural network to train and optimise the ranging data gathered in the 1–20 m range in order to solve the issue of significant ranging errors in conventional UWB positioning systems. The experimental data indicates that the UWB positioning system’s unoptimized range error has an absolute mean of 0.18 m and a standard deviation of 0.047. However, when using a neural network-trained model, the ranging error is much decreased, with an absolute mean of 0.038 m and a standard deviation of 0.0079. The average root mean square error (RMSE) of the positioning coordinates is decreased to 0.043 m following the positioning computation utilising the optimised range data, greatly increasing the positioning accuracy. This study used the conventional camera shooting method for image acquisition. Following image acquisition, the system extracts the cow’s coordinate information from the image using a perspective transformation method. This allows for accurate cow identification and number labelling when compared to the location coordinates. According to the trial findings, this plan, which integrates computer vision and UWB positioning technologies, achieves high-precision cow labelling and placement in the optimised system and greatly raises the degree of automation and precise management in the farming process. This technology has many potential applications, particularly in the administration and surveillance of big dairy farms, and it offers a strong technical basis for precision farming.

A Data-Driven Approach for Automatic Aircraft Engine Borescope Inspection Defect Detection Using Computer Vision and Deep Learning

Regular aircraft engine inspections play a crucial role in aviation safety. However, traditional inspections are often performed manually, relying heavily on the judgment and experience of operators. This paper presents a data-driven deep learning framework capable of automatically detecting defects on reactor blades. Specifically, this study develops Deep Neural Network models to detect defects in borescope images using various datasets, based on Computer Vision and YOLOv8n object detection techniques. Firstly, reactor blade images are collected from public resources and then annotated and preprocessed into different groups based on Computer Vision techniques. In addition, synthetic images are generated using Deep Convolutional Generative Adversarial Networks and a manual data augmentation approach by randomly pasting defects onto reactor blade images. YOLOv8n-based deep learning models are subsequently fine-tuned and trained on these dataset groups. The results indicate that the model trained on wide-shot blade images performs better overall at detecting defects on blades compared to the model trained on zoomed-in images. The comparison of multiple models’ results reveals inherent uncertainties in model performance that while some models trained on data enhanced by Computer Vision techniques may appear more reliable in some types of defect detection, the relationship between these techniques and subsequent results cannot be generalized. The impact of epochs and optimizers on the model’s performance indicates that incorporating rotated images and selecting an appropriate optimizer are key factors for effective model training. Furthermore, models trained solely on artificially generated images from collages perform poorly at detecting defects in real images. A potential solution is to train the model on both synthetic and real images. Future work will focus on improving the framework’s performance and conducting a more comprehensive uncertainty analysis by utilizing larger and more diverse datasets, supported by enhanced computational power.

Real-time digital dermatitis detection in dairy cows on Android and iOS apps using computer vision techniques

Abstract The aim of the study was to deploy computer vision (CV) models for real-time detection of digital dermatitis (DD) lesions in cows using Android or iOS mobile applications. Early detection of DD lesions in dairy cows is crucial for prompt treatment and animal welfare. Android and iOS apps could facilitate routine and early DD detection in cows’ feet on dairy and beef farms. Upon detecting signs of DD, dairy farmers could implement preventive and treatment methods, including foot baths, topical treatment, hoof trimming, or quarantining cows affected by DD to prevent its spread. We applied transfer-learning to DD image data for 5 lesion classes, M0, M4H, M2, M2P, and M4P, on pretrained YOLOv5 model architecture using COCO-128 pretrained weights. The combination of localization loss, classification loss, and objectness loss was used for the optimization of prediction performance. The custom DD detection model was trained on 363 images of size 416 x 416 pixels and tested on 46 images. During model training, data were augmented to increase model robustness in different environments. The model was converted into TFLite format for Android devices and CoreML format for iOS devices. Techniques such as quantization were implemented to improve inference speed in real-world settings. The DD models achieved an average mAP (mean Average Precision) of 0.95 on the test dataset. When tested in real-time, iOS devices resulted in Cohen’s kappa value of 0.57 (95% CI: 0.49-0.65) averaged across the 5 lesion classes denoting the moderate agreement of the model detection with human investigators. The Android device resulted in a Cohen’s kappa value of 0.38 (95% CI: 0.29-0.47) denoting fair agreement between model and investigator. Combining M2 and M2P classes and M4H and M4P classes resulted in a Cohen’s kappa value of 0.65 (95% CI: 0.54-0.76) and 0.46 (95% CI: 0.35-0.57), for Android and iOS devices, respectively. For the two-class model (lesion vs. non-lesion), a Cohen’s kappa value of 0.74 (95% CI: 0.63-0.85) and 0.65 (95% CI: 0.52-0.78) was achieved for iOS and Android devices, respectively. iOS achieved a good inference time of 20 ms, compared to 57 ms on Android. Additionally, we deployed models on Ultralytics iOS and Android apps giving kappa scores of 0.56 (95% CI: 0.48-0.64) and 0.46 (95% CI: 0.37-0.55) respectively. Our custom iOS app surpassed the Ultralytics apps in terms of kappa score and the confidence score.

A Self-Adaptive Traffic Signal System Integrating Real-Time Vehicle Detection and License Plate Recognition for Enhanced Traffic Management

Traffic management systems play a crucial role in smart cities, especially because increasing urban populations lead to higher traffic volumes on roads. This results in increased congestion at intersections, causing delays and traffic violations. This paper proposes an adaptive traffic control and optimization system that dynamically adjusts signal timings in response to real-time traffic situations and volumes by applying machine learning algorithms to images captured through video surveillance cameras. This system is also able to capture the details of vehicles violating signals, which would be helpful for enforcing traffic rules. Benefiting from advancements in computer vision techniques, we deployed a novel real-time object detection model called YOLOv11 in order to detect vehicles and adjust the duration of green signals. Our system used Tesseract OCR for extracting license plate information, thus ensuring robust traffic monitoring and enforcement. A web-based real-time digital twin complemented the system by visualizing traffic volume and signal timings for the monitoring and optimization of traffic flow. Experimental results demonstrated that YOLOv11 achieved a better overall accuracy, namely 95.1%, and efficiency compared to previous models. The proposed solution reduces congestion and improves traffic flow across intersections while offering a scalable and cost-effective approach for smart traffic and lowering greenhouse gas emissions at the same time.

Drowsiness Detection of Construction Workers: Accident Prevention Leveraging Yolov8 Deep Learning and Computer Vision Techniques

Drowsiness Detection of Construction Workers: Accident Prevention Leveraging Yolov8 Deep Learning and Computer Vision Techniques

Advancing Source Apportionment of Atmospheric Particles: Integrating Morphology, Size, and Chemistry Using Electron Microscopy Technology and Machine Learning.

To further reduce atmospheric particulate matter concentrations, there is a need for a more precise identification of their sources. The SEM-EDS technology (scanning electron microscopy and energy-dispersive X-ray spectroscopy) can provide high-resolution imaging and detailed compositional analysis for particles with relatively stable physical and chemical properties. This study introduces an advanced source apportionment pipeline (RX model) that uniquely combines computer-controlled scanning electron microscopy with computer vision and machine learning to trace particle sources by integrating single-particle morphology, size, and chemical information. In the evaluation using a virtual data set with known source contributions, the RX model demonstrated high accuracy, with average errors of 0.60% for particle number and 1.97% for mass contribution. Compared to the chemical mass balance model, the RX model's accuracy and stability improved by 75.6 and 73.4%, respectively, and proved effective in tracing Fe-containing particles in the atmosphere of a steel city in China. This study indicates that particle morphology can serve as an effective feature for determining its source. The findings highlight the potential of electron microscopy technology coupled with computer vision and machine learning techniques to enhance our understanding of atmospheric pollution sources, offering valuable insights for PM health risk assessment and evidence-based policy-making.

Estimating the legibility of international borders.

Researchers in the social sciences are interested in the consequences of institutions, increasingly on a global scale. Institutions that may be negotiated between states can have consequences at a microlevel, as local populations adjust their expectations and ultimately even their behavior to take institutional rules into account. However, large-scale fine-grained analyses that test for the complex evidence of such institutions locally are rare. This article focuses on a key institution: International borders. Using computer vision techniques, we show that it is possible to produce a geographically specific, validated, and replicable way to characterize border legibility, by which we mean the ability to visually detect the presence of an international border in physical space. We develop and compare computer vision techniques to automatically estimate legibility scores for 627,656 imagery tiles from virtually every border in the world. We evaluate statistical and data-driven computer vision methods, finding that fine-tuning pretrained visual recognition models on a small set of human judgments allows us to produce local legibility scores globally that align well with human notions of legibility. Finally, we interpret these scores as useful approximations of states' border orientations, a concept that prior literature has used to capture the visible investments states make in border areas to maintain jurisdictional authority territorially. We validate our measurement strategy using both human judgments and five nomological validation indicators.

SBCP-YOLO-R3D: Student Behavior Recognition and Visualization Framework Using Improved YOLO and R3D for Class Video

Real-time recognition and visualization of students' behaviors in face-to-face classrooms serve as pivotal indicators of learning engagement. However, current methods exhibit limitations in both real-time performance and accuracy. Additionally, in-depth studies have not been extensively conducted to evaluate learning status more conveniently by utilizing computer vision techniques. To address these issues, a novel Student Behavior Recognition and Dynamic Class Portraits Construction framework, named SBCP-YOLO-R3D, incorporating the StB-YOLO and R3D methods, has been proposed to detect student behaviors and construct class portraits. The developed framework comprises two layers: the StB-YOLO detection layer and the R3D classification layer. In the StB-YOLO detection layer, the Lightweight-SEAM (LW-SEAM) is incorporated into YOLOv5 to enhance the recognition of occluded students, by capturing contextual information and enhancing occlusion-related features. Moreover, a Double-SlideLoss function is devised, employing adaptive weighting mechanisms to strike an optimal balance between simple and challenging samples. In the R3D classification layer, the results generated by StB-YOLO are then processed using R3D to produce class portraits. Experiments conducted on the StuAct and SCB-DATASET3-S datasets demonstrate the effectiveness of the StB-YOLO. Compared with the baseline model, StB-YOLO increases the mAP by 3.1%.

Reducing the social inequity of neighborhood visual environment in Los Angeles through computer vision and multi-model machine learning.

Aligning with the United Nations' Sustainable Development Goals, the focus on creating safe, sustainable cities and enhancing the wellbeing of individuals across all age groups has become a central aspect of urban planning and environmental management. The environments we live in significantly influence our thoughts, emotions, and interactions with the world around us. Our study aims to unveil the social inequity of citizens' wellbeing, reflected by their perception on neighborhood environment (e.g., feeling of depression), across different social/vulnerable groups (i.e., White, Black, Asian, Hispanic, low-income, low-educated, and unemployed) via crowdsourced street view imageries and computer vision. Specifically, we quantified the actual built environment in the 5D dimensions (i.e., density, diversity, design, distance, and destination) based on multiple sources; measured six types of neighborhood visual environment (i.e., perception of beautiful, safe, wealthy, liveable, boring and depressing) and the overall neighborhood soundness index by using computer vision technique and street view imageries collected from Mapillary; and unveiled the actual built environmental features that are associated with people's visual perception towards the surrounding environment via multi-model machine learning methods. Our pilot study in Los Angeles County finds that neighborhoods with higher concentrations of Black, Hispanic, low-income, low-educated, and unemployed populations are perceived as less beautiful, liveable, safe, and wealthy. The most important actual built environment features positively influencing human perception include the density of canopy, followed by the density of multiple units, the distance to CBD, and car commuting to destinations, regardless of social groups. Our key findings provide place-based evidence for the design and upgrading of the community environment that further affects people's daily activity and living style. Our framework and methods can be applied to cross-disciplinary studies, aiding urban planning and healthy city initiatives with place-based evidence.

6-DoF Pose Estimation from Single RGB Image and CAD Model Retrieval Using Feature Similarity Measurement

This study presents six degrees of freedom (6-DoF) pose estimation of an object from a single RGB image and retrieval of the matching CAD model by measuring the similarity between RGB and CAD rendering images. The 6-DoF pose estimation of an RGB object is one of the important techniques in 3D computer vision. However, in addition to 6-DoF pose estimation, retrieval and alignment of the matching CAD model with the RGB object should be performed for various industrial applications such as eXtended Reality (XR), Augmented Reality (AR), robot’s pick and place, and so on. This paper addresses 6-DoF pose estimation and CAD model retrieval problems simultaneously and quantitatively analyzes how much the 6-DoF pose estimation affects the CAD model retrieval performance. This study consists of two main steps. The first step is 6-DoF pose estimation based on the PoseContrast network. We enhance the structure of PoseConstrast by adding variance uncertainty weight and feature attention modules. The second step is the retrieval of the matching CAD model by an image similarity measurement between the CAD rendering and the RGB object. In our experiments, we used 2000 RGB images collected from Google and Bing search engines and 100 CAD models from ShapeNetCore. The Pascal3D+ dataset is used to train the pose estimation network and DELF features are used for the similarity measurement. Comprehensive ablation studies about the proposed network show the quantitative performance analysis with respect to the baseline model. Experimental results show that the pose estimation performance has a positive correlation with the CAD retrieval performance.

"Play by play": A dataset of handball and basketball game situations in a standardized space.

This paper presents a synthetic dataset of labeled game situations in recordings of federated handball and basketball matches played in Galicia, Spain. The dataset consists of synthetic data generated from real video frames, including 308,805 labeled handball frames and 56,578 labeled basketball frames extracted from 2105 handball and 383 basketball 5-s video clips. Experts manually labeled the video clips based on the respective sports, while the individual frames were automatically labeled using computer vision and machine learning techniques. The dataset encompasses seven classes of game situations: left attack, left counterattack, left penalty, right attack, right counterattack, right penalty, and timeout. In basketball, the penalty class refers to the free throws attempted by players after they have been fouled by an opposing player. Each frame in the dataset is assigned to one of these classes, considering the game situation and specific context. Importantly, the dataset does not contain actual video frames; instead, it provides a synthetic, normalized representation of each frame in JSON format. This tabular data includes player, referee, and ball positions on a normalized field, player and referee velocities, and key regions on the court. Positions of players, referees, and the ball were automatically inferred in each frame by an object detector, followed by a tracking step to detect object positions across frames and compute the velocity vectors. Finally, the obtained coordinates underwent normalization through a perspective transformation, ensuring that the data remained unaffected by variations in camera configurations across different arenas and camera setups. We refer to this standardized coordinate space as the 'unified space'. The dataset holds significant potential for reuse in various domains related to sports analytics and machine learning research. It can serve as a valuable resource for researchers, coaches, and sports enthusiasts, contributing to improvements in player performance, game strategies, match retransmissions, and sports-related technologies.

Recent advancements and future directions in automatic swallowing analysis via videofluoroscopy: A review.

Videofluoroscopic swallowing studies (VFSS) capture the complex anatomy and physiology contributing to bolus transport and airway protection during swallowing. While clinical assessment of VFSS can be affected by evaluators subjectivity and variability in evaluation protocols, many efforts have been dedicated to developing methods to ensure consistent measures and reliable analyses of swallowing physiology using advanced computer-assisted methods. Latest advances in computer vision, pattern recognition, and deep learning technologies provide new paradigms to explore and extract information from VFSS recordings. The literature search was conducted on four bibliographic databases with exclusive focus on automatic videofluoroscopic analyses. We identified 46 studies that employ state-of-the-art image processing techniques to solve VFSS analytical tasks including anatomical structure detection, bolus contrast segmentation, and kinematic event recognition. Advanced computer vision and deep learning techniques have enabled fully automatic swallowing analysis and abnormality detection, resulting in improved accuracy and unprecedented efficiency in swallowing assessment. By establishing this review of image processing techniques applied to automatic swallowing analysis, we intend to demonstrate the current challenges in VFSS analyses and provide insight into future directions in developing more accurate and clinically explainable algorithms.

Android Based Visual Product Identification System for the Blind

The proposed system, "Android-Based Visual Product Identification for the Blind," aims to enhance the independence of visually impaired individuals by enabling them to identify everyday products using an Android smartphone. The system leverages computer vision techniques integrated with Android’s mobile platform to capture images of products, analyse them using a trained machine learning model, and provide audio feedback to the user. The application focuses on product packaging, barcodes, and distinct visual features, ensuring accurate identification in real-time. The solution is designed to be user-friendly with voice commands, accessible navigation, and low latency, making it practical for daily use. By offering a portable and reliable means of product identification, this tool addresses the challenges faced by the blind in interacting with their surroundings, improving their quality of life and fostering independence.

A blind navigation guide model for obstacle avoidance using distance vision estimation based YOLO-V8n

Obstacle is an object positioned along a path of propagation with the potential to cause a collision and hence, an accident. Over the years, several papers have applied advanced computer vision techniques, particularly transfer learning algorithms, to solve this problem, but despite their success, in specific vision applications such as blind guide navigation systems, the model finds it difficult to distinguish between objects and obstacles recognized in the same video frame, hence attracting research attention. In this paper, the aim was to develop a blind navigation guide model for obstacle avoidance using distance vision estimation-based YOLO-V8n. To achieve this, an improved data model was developed using the MS COCO dataset and primary data collected from several indoor environments. Then, the YOLO-V8n architecture was improved by adding a Weighted Feature Enhancement (WFE) model to the backbone for improved feature extraction, and Bi-directional Feature Pyramid Network (Bi-FPN) was applied to the neck to improve multi-scale feature representation. In addition, a Distance Vision Estimation (DVE) algorithm was developed and applied to the Bi-FPN before connecting it to the head of the YOLO-V8n to facilitate simultaneous object detection and distance measurement in real-time video. Furthermore, the issue of bounding box overlap in the model was addressed by applying a Wise Intersection over Unit (WIoU) loss function. Collectively, these formulated the new transfer learning algorithm called YOLO-V8n+WFE+Bi-FPN+DVE+WIoU used in this work for high-level obstacle detection and distance estimation. The model was trained considering different experimental architectures of the YOLO-V8 and loss functions, respectively, and then evaluated with precision, recall, mean absolute precision, and average precision, respectively, before validation through comparative analysis. Upon selection of the best model, it was further validated through comparison with other state-of-the art algorithms before deployment for obstacle avoidance in an indoor environment, having satisfied the condition of reliability. Real world testing of the model was performed at four different indoor sites, and the results showed that while the model was able to correctly classify objects, it could also measure their distance accurately, thereby making it suitable for deployment as a blind vision guide navigation system.

A novel hybrid inception-xception convolutional neural network for efficient plant disease classification and detection

Plants are essential at all stages of living things. Plant pests, diseases, and symptoms are most regularly visible in plant leaves and fruits and sometimes within the roots. Yet, their diagnosis by experts in the laboratory is expensive, tedious, and time-consuming if the samples involve laboratory analysis. Failure to detect early plant symptoms and diseases is the core biotic cause of increased plant stresses, structure, health, reduced subsistence farming, and threats to global food security. To mitigate these problems at a social, economic, and environmental level, inappropriate herbicide application reduction and early plant disease detection and classification (PDDC) are significant solutions in this case. Advancements in transfer learning techniques have resulted in effective results in smart farming and have become extensively used in disease identification and classification research studies. This study presents a novel hybrid inception-xception (IX) using a convolution neural network (CNN). The presented model combines inception and depth-separable convolution layers to capture multiple-scale features while reducing model complexity and overfitting. In contrast to ordinary CNN architectures, it extends the network for better feature extraction, improving PDDC performance that demands diverse feature competencies. It further presents a real-time artificial intelligence (AI) application available in MATLAB, Android, and Servlet to automatically identify and classify diseases based on the leaf environment using improved CNN, machine learning (ML), and computer vision techniques. To assess the presented IX-CNN model performance, different classifiers, namely, support vector machine (SVM), decision tree (DT) and random forest (RF), were used. The experiments used six datasets, including PlantVillage, Turkey Disease, Plant Doc, Rice Disease, RoCole, and NLB datasets. Plant Doc, PlantVillage, and Turkey Disease datasets demonstrated an accuracy of 100%. Rice Disease, RoCole, and NLB attained an accuracy of 99.79%, 99.95%, and 98.64%, respectively.

Detection of Military Aircraft Using YOLO and Transformer-Based Object Detection Models in Complex Environments

Computer vision and deep learning techniques are widely applied in object detection tasks across various domains, including defense technologies. Accurate and efficient detection of military aircraft plays a critical role in strengthening air defense systems and enabling effective strategic decision-making. This study evaluates the performance of YOLOv7, YOLOv8, and RT-DETR models in detecting military aircraft using a dataset consisting of 19.514 images spanning 43 aircraft models. The dataset incorporates images captured from various angles and diverse backgrounds, such as urban, rural, and coastal areas, ensuring realistic testing conditions. However, class imbalance is observed, with certain aircraft models, such as the F14 and F16, being more represented than others, which may affect model generalization. To address these challenges, hyperparameters were optimized, and performance metrics, including mean Average Precision (mAP) and recall, were analyzed. Experimental results show that YOLOv8 achieved 94% mAP and 88.1% recall, YOLOv7 reached 90.2% mAP and 82.7% recall, while RT-DETR demonstrated consistent performance with 92.7% mAP and 90.4% recall. These findings highlight the strengths and limitations of the evaluated models and provide inferences for improving detection systems in defense applications.

A Systematic Literature Review of Vision-Based Fire Detection, Prediction and Forecasting

The primary method used by conventional fire detection systems is sensor-based detection, which has limitations in terms of accuracy and detection time. Traditional approaches and techniques could be improved by the latest advancements in computer vision-based technologies for fire prediction and detection. Consequently, this paper aims to provide a comprehensive literature analysis of earlier research on fire detection and prediction using the computer vision techniques. The Preferred Reporting Items for Systemic Reviews and Meta-Analyses, or PRISMA 2020, are applied in this systematic review. Three databases such as the Web of Science, Scopus, and IEEE were searched for pertinent publications to include in this review for this study. The systematic review reveals that existing studies predominantly focused on fire flame rather than smoke detection. Moreover, the majority of research has centered on forest fires in the particular context of occurrence, neglecting indoor or interior environments. Video surveillance systems emerge as the primary source of hardware and datasets utilized in these investigations. Notably, convolutional neural networks (CNNs) stand out as the most frequently employed deep learning approach for classification purposes. The systematic review clarifies the state of fire detection research using computer vision techniques by combining data from several academic sources. Through a systematic approach, this study contributes a deeper understanding of the opportunity and challenges in leveraging visionbased technologies for fire detection and prediction.

Convolutional Neural Network for Depression and Schizophrenia Detection

Background/Objectives: This study presents a Convolutional Neural Network (CNN) approach for detecting depression and schizophrenia using motor activity patterns represented as images. Participants’ motor activity data were captured and transformed into visual representations, enabling advanced computer vision techniques for the classification of these mental disorders. The model’s performance was evaluated using a three-fold cross-validation, achieving an average accuracy of 95%, demonstrating the effectiveness of the proposed approach in accurately identifying mental health conditions. The objective of the study is to develop a model capable of identifying different mental disorders by processing motor data using CNN in order to provide a support tool to areas specialized in the diagnosis and efficient treatment of these psychological conditions. Methods: The methodology involved segmenting and transforming motor activity data into images, followed by a CNN training and testing phase on these visual representations. This innovative approach enables the identification of subtle motor behavior patterns, potentially indicative of specific mental states, without invasive interventions or self-reporting. Results: The results suggest that CNNs can capture discriminative features in motor activity to differentiate between individuals with depression, schizophrenia, and those without mental health diagnoses. Conclusions: These findings underscore the potential of computer vision and deep neural network techniques to contribute to early, non-invasive mental health disorder diagnosis, with significant implications for developing mental health support tools.