Single Camera Research Articles

Optimizing Camera Setup for In-Home First-Person Rendering Mixed-Reality Gaming System

With the advent of 3D humanoid reconstruction techniques, using a realistic 3D human avatar in a serious game has become popular. This realistic representation in the virtual environment could be achieved using a single low-cost RGB-D camera such as Kinect. However, properly setting up such a camera system for high-quality rendering can be challenging due to the relatively restricted in-home environment. In this paper, we address the challenge of finding optimized camera setup guidelines for an in-home first-person perspective mixed reality (MR) gaming system. We use an MR system with personalized humanoids to simulate the texture reconstruction for a user under a specific camera configuration. Then, a derivative-free optimization is leveraged as a black box approach to search for the optimized camera setup through iterative simulation. We also introduce a novel skeleton-based calibration to address the effects of physically varying the camera's position. For evaluation, two experiments are carried out to evaluate the correctness and effectiveness of the proposed calibration. Furthermore, we conduct a case study using simulation-based optimization for reconstructing lower limb amputees. This work can potentially help locate a proper camera setup for an MR system within the constraints of an in-home environment.

Characterization of a SiPM-based monolithic neutron scatter camera using dark counts

The Single Volume Scatter Camera (SVSC) Collaboration aims to develop portable neutron imaging systems for a variety of applications in nuclear non-proliferation. Conventional double-scatter neutron imagers are composed of several separate detector volumes organized in at least two planes. A neutron must scatter in two of these detector volumes for its initial trajectory to be reconstructed. As such, these systems typically have a large footprint and poor geometric efficiency. We report on the design and characterization of a prototype monolithic neutron scatter camera that is intended to significantly improve upon the geometrical shortcomings of conventional neutron cameras. The detector consists of a 50 mm×56 mm× 60 mm monolithic block of EJ-204 plastic scintillator instrumented on two faces with arrays of 64 Hamamatsu S13360-6075PE silicon photomultipliers (SiPMs). The electronic crosstalk is limited to < 5% between adjacent channels and < 0.1% between all other channel pairs. SiPMs introduce a significantly elevated dark count rate over PMTs, as well as correlated noise from after-pulsing and optical crosstalk. In this article, we characterize the dark count rate and optical crosstalk and present a modified event reconstruction likelihood function that accounts for them. We find that the average dark count rate per SiPM is 4.3 MHz with a standard deviation of 1.5 MHz among devices. The analysis method we employ to measure internal optical crosstalk also naturally yields the mean and width of the single-electron pulse height. We calculate separate contributions to the width of the single-electron pulse-height from electronic noise and avalanche fluctuations. We demonstrate a timing resolution for a single-photon pulse to be (128 ± 4) ps. Finally, coincidence analysis is employed to measure external (pixel-to-pixel) optical crosstalk. We present a map of the average external crosstalk probability between 2×4 groups of SiPMs, as well as the in-situ timing characteristics extracted from the coincidence analysis. Further work is needed to characterize the performance of the camera at reconstructing single- and double-site interactions, as well as image reconstruction.

Spatial Characterization of Multiple Pulses, With Different Spatial Profiles, From a Single Camera Snapshot

Spatial Characterization of Multiple Pulses, With Different Spatial Profiles, From a Single Camera Snapshot

Offistretch: camera-based real-time feedback for daily stretching exercises

AbstractIn this paper, we present OffiStretch, a camera-based system for optimal stretching guidance at home or in the workplace. It consists of a vision-based method for real-time assessment of the user’s body pose to provide visual feedback as interactive guidance during stretching exercises. Our method compares the users’ actual pose with a pre-trained target pose to assess the quality of stretching for a number of different exercises. We utilize angular and spatial pose features to perform this comparison for each individual exercise. The result of this pose assessment is presented to the user as real-time visual feedback on an "augmented mirror" display. As our method relies simply on a single RGB camera, it can be easily utilized in everyday training scenarios. We validate our method in a user study, comparing users’ performance and motivation in stretching when receiving audio-visual guidance on a TV screen both with and without our live feedback. While participants performed equally well in both conditions, feedback boosted their motivation to perform the exercises, highlighting its potential for increasing users’ well-being. Moreover, our results suggest that participants preferred stretching exercises with our live feedback over the condition without the feedback. Finally, an expert evaluation with professional physiotherapists reveals that further work must target improvements of the feedback to ensure correct guidance during stretching.

Clustered and rotating designs as a strategy to obtain precise detection rates in camera trapping studies

Abstract Camera traps have transformed the way we monitor wildlife and are now routinely used to address questions from a wide range of ecological and conservation aspects. Sampling design optimization and a better understanding of drivers determining the precision of detection rates (i.e. the number of detections per unit of effort) are important methodological issues. Little attention has been focused on the effect of placing more than one camera on each sampling point (hereafter, clustered design), and/or rotating (i.e. redeploying) the cameras to new placements during the sampling period. We explored the differences in the precision of detection rates between clustered vs. single camera designs when cameras remained in the same location during the study. Furthermore, the effect of keeping the placement of cameras fixed or rotating them (i.e. moving them to new locations during the sampling period), when a limited number of camera devices are available, was also evaluated. We used simulations and field data to test differences in detection rate precision for the different sampling designs. We simulated three different population distributions (random, trail‐based and aggregated) and three abundance scenarios. The simulations were validated with a field experiment focused on eight species with different behavioural traits, including artiodactyls, carnivores, lagomorphs, and birds. When a fixed number of sampling points were monitored simultaneously, clustered designs generally resulted in an increase in the precision of detection rates compared to single designs. The absolute reduction in the coefficient of variation by clustered designs was on average 0.07 units (min: 0.01, max: 0.15), which represents an average relative reduction in CV of 31% (min:6%, max:44%). An improvement in precision was also observed as a higher number of sampling points was used for all population distributions and sampling designs tested. When a fixed number of cameras were available, rotating the cameras to independent locations improved precision (an absolute reduction of 0.19 CV units) when monitoring aggregated populations, but not for random and trail‐based population distributions. Synthesis and applications: Our research provides a guideline for wildlife managers and researchers to improve the precision of camera trap detection rates and optimize resource allocation. In general, the study design should accommodate the behaviour of the target species (e.g. spatial aggregation and abundance), monitoring program logistic resources (both human and economic) and study area characteristics (e.g. accessibility and vandalism).

Single-exposure multi-wavelength optical diffraction tomography based on space-angle dual multiplexing holography.

We present a space-angle dual multiplexing holographic recording system for realizing single-exposure multi-wavelength optical diffraction tomographic (ODT) imaging. This system is achieved by combining the principle of single-exposure multi-wavelength holographic imaging technique based on angle-division multiplexing with the principle of single-exposure ODT imaging technique based on microlens array multi-angle illuminations and space-division multiplexing. Compared with the existing multi-wavelength ODT imaging methods, it enables the holographic recording of all the diffraction tomography information of a measured specimen at multiple illumination wavelengths in a single camera exposure without any scan mechanism. Using our proposed data processing method, the multi-wavelength three-dimensional (3D) refractive index tomograms of a specimen can be eventually reconstructed from single recorded multiplexing hologram. Experimental results of a static polystyrene bead and a living C. elegans worm demonstrate the feasibility of this system.

Multi-View Metal Parts Pose Estimation Based on a Single Camera.

Pose estimation of metal parts plays a vital role in industrial grasping areas. It is challenging to obtain complete point clouds of metal parts because of their reflective properties. This study introduces an approach for recovering the 6D pose of CAD-known metal parts from images captured by a single RGB camera. The proposed strategy only requires RGB images without depth information. The core idea of the proposed method is to use multiple views to estimate the metal parts' pose. First, the pose of metal parts is estimated in the first view. Second, ray casting is employed to simulate additional views with the corresponding status of the metal parts, enabling the calculation of the camera's next best viewpoint. The camera, mounted on a robotic arm, is then moved to this calculated position. Third, this study integrates the known camera transformations with the poses estimated from different viewpoints to refine the final scene. The results of this work demonstrate that the proposed method effectively estimates the pose of shiny metal parts.

Inclusion of a skeletal model partly improves the reliability of lower limb joint angles derived from a markerless depth camera

A single depth camera provides a fast and easy approach to performing biomechanical assessments in a clinical setting; however, there are currently no established methods to reliably determine joint angles from these devices. The primary aim of this study was to compare joint angles as well as the between-day reliability of direct kinematics to model-constrained inverse kinematics recorded using a single markerless depth camera during a range of clinical and athletic movement assessments.A secondary aim was to determine the minimum number of trials required to maximize reliability. Eighteen healthy participants attended two testing sessions one week apart. Tasks included treadmill walking, treadmill running, single-leg squats, single-leg countermovement jumps, bilateral countermovement jumps, and drop vertical jumps. Keypoint data were processed using direct kinematics as well as in OpenSim using a full-body musculoskeletal model and inverse kinematics. Kinematic methods were compared using statistical parametric mapping and between-day reliability was calculated using intraclass correlation coefficients, mean absolute error, and minimal detectable change. Keypoint-derived inverse kinematics resulted in significantly smaller hip flexion (range = −9 to −2°), hip abduction (range = −3 to −2°), knee flexion (range = −5° to −2°), and greater dorsiflexion angles (range = 6–15°) than direct kinematics. Both markerless kinematic methods had high between-day reliability (inverse kinematics ICC 95 %CI = 0.83–0.90; direct kinematics ICC 95 %CI = 0.80–0.93). For certain tasks and joints, keypoint-derived inverse kinematics resulted in greater reliability (up to 0.47 ICC) and smaller minimal detectable changes (up to 13°) than direct kinematics. Performing 2–4 trials was sufficient to maximize reliability for most tasks. A single markerless depth camera can reliably measure lower limb joint angles, and skeletal model-constrained inverse kinematics improves lower limb joint angle reliability for certain tasks and joints.

Fall Detection and Boundary Detection in Care Homes

The elderly population represents a significant and rapidly expanding demographic, with a majority experiencing frequent daily accidents, notably falls. Falls rank as the second leading cause of accidental injury deaths globally. To address this issue, we propose a video classification system designed specifically for fall detection. Our fall detection framework comprises two key steps: firstly, the detection of human posture within video frames, followed by fall classification using Convolutional Neural Networks (CNNs). Additionally, we introduce a novel approach for boundary detection, utilizing object detection techniques beyond a predefined line of surveillance captured by a single camera. Through this integrated methodology, we aim to enhance fall detection and boundary breach detection capabilities, thereby contributing to the advancement of elderly care and safety. (Abstract).

Route Positioning System for Campus Shuttle Bus Service Using a Single Camera

A route positioning system is a technology that identifies the current route when driving from one stop to the next, commonly found in public transportation systems such as shuttle buses that follow fixed routes. This is especially useful for smaller-scale services, such as shuttle buses, where using expensive technology and sensors for location tracking might not be feasible. Particularly in urban areas with tall buildings or mountainous regions with lots of trees, relying solely on GPS can lead to many errors. Therefore, this paper suggests a cost-effective solution that uses just one camera sensor to accurately determine the location of small-scale transportation services on fixed routes. For this, this paper uses a single-stage detection network that quickly identifies objects and tracks them using a simple algorithm. These detected features are compiled into a “codebook” using the bag-of-visual-words technique. During actual trips, this pre-created codebook is compared with landmarks that the camera sees. This comparison helps to determine the route currently being traveled. To test the effectiveness of this approach, this paper used the route of a shuttle bus on the Gachon University campus, which is similar to a downtown area with tall buildings or a wooded mountainous area. The results showed that the shuttle bus’s route was recognized with an accuracy of 0.60. Areas with distinct features were recognized with an accuracy of 0.99, while stops with simple, nondescript structures were recognized with an accuracy of 0.29. Additionally, applying the SORT algorithm to enhance performance slightly improved the accuracy from 0.60 to 0.61. This demonstrates that our proposed method can effectively perform location recognition using only cameras in small shuttle buses.

Single shot depth-resolved imaging through dynamic turbid media

Guide star assisted wavefront shaping techniques have been exploited for focusing and imaging through turbid media by addressing a scatter inverse pattern. However, the turbid medium is required to be steady before finding the proper correction pattern, which limits applications in focusing and imaging through dynamic media, such as turbid water or blood. This study proposes a holography-based dual-polarization computational wavefront shaping method for imaging objects at variant depths behind dynamic turbid media. The orthogonal polarized output speckles of a point source (considered as a guide star) and an object are simultaneously recorded in holograms in separate regions of a single CCD camera. The holograms of the point source and object are subjected to the same distortion regardless of whether the media is static or dynamic. The hologram of the point source is used to determine the correction phase pattern for the distortion, while that of the object is used to record the complex scattered wavefront of the object. To reconstruct a clear object image, the wavefront of the scattered object is digitally corrected using the correction phase pattern and is then transferred to the image plane by calculating the transmission of the angular spectrum. Benefiting from the autofocusing feature of digital holography, objects at different depths can be recovered from a single shot hologram pair. The potential applications of the proposed method in diverse dynamic scattering media are demonstrated by imaging through a moving diffuser, turbid water, and pig blood with optical depth beyond 10.

Improved markerless gait kinematics measurement using a biomechanically-aware algorithm with subject-specific geometric modeling

Despite the advancements in developing markerless gait analysis systems, they still demonstrate lower accuracy compared to gold-standard systems. Hence, in this research, a novel approach is presented to improve the lower limb kinematics accuracy in markerless gait analysis. This approach refines the 3D lower-limb skeletons obtained by AI-based pose estimation algorithms in a subject-specific geometric manner, preserves skeleton links’ length, benefits from gait phases information that adds biomechanical awareness to the algorithm, and utilizes an embedded trajectory smoothing. Validation of the proposed method shows that it reduces 12.6–43.5 % of root mean square error (RMSE) and significantly improves kinematic curves’ similarity to the gold-standard ones. Results also prove the feasibility of more accurate lower limb kinematics calculation using a single (2.02–7.57° RMSE) or dual RGB-D camera (1.66–7.25° RMSE). Development of such algorithms could result in requirement of fewer cameras that deliver comparable or even superior measurement accuracy compared to multi-camera approaches.

Real-Time Seamless Multi-Projector Displays on Deformable Surfaces.

Prior works on multi-projector displays have focused primarily on static rigid objects, some focusing on dynamic rigid objects. However, works on projection based displays on deformable dynamic objects have focused only on small scale single projector displays. Tracking a deformable dynamic surface and updating projections precisely in real time on it is a significantly challenging task, even for a single projector system. In this paper, we present the first end-to-end solution for achieving a real-time, seamless display on deformable surfaces using mutliple unsychronized projectors without requiring any prior knowledge of the surface or device parameters. The system first accurately calibrates multiple RGB-D cameras and projectors using the deformable display surface itself, and then using those calibrated devices, tracks the continuous changes in the surface shape. Based on the deformation and projector calibration, the system warps and blends the image content in real-time to create a seamless display on a surface that continuously changes shape. Using multiple projectors and RGB-D cameras, we provide the much desired aspect of scale to the displays on deformable surfaces. Most prior dynamic multi-projector systems assume rigid objects and depend critically on the constancy of surface normals and non-existence of local shape deformations. These assumptions break in deformable surfaces making prior techniques inapplicable. Point-based correspondences become inadequate for calibration, exacerbated with no synchronization between the projectors. A few works address non-rigid objects with several restrictions like targeting semi-deformable surfaces (e.g. human face), or using single coaxial (optically aligned) projector-camera pairs, or temporally synchronized cameras. We break loose from such restrictions and handle multiple projector systems for dynamic deformable fabric-like objects using temporally unsynchronized devices. We devise novel methods using ray and plane-based constraints imposed by the pinhole camera model to address these issues and design new blending methods dependent on 3D distances suitable for deformable surfaces. Finally, unlike all prior work with rigid dynamic surfaces that use a single RGB-D camera, we devise a method that involve all RGB-D cameras for tracking since the surface is not seen completely by a single camera. These methods enable a seamless display at scale in the presence of continuous movements and deformations. This work has tremendous applications on mobile and expeditionary systems where environmentals (e.g. wind, vibrations, suction) cannot be avoided. One can create large displays on tent walls in remote, austere military or emergency operations in minutes to support large scale command and control, mission rehearsal or training operations. It can be used to create displays on mobile and inflatable objects for tradeshows/events and touring edutainment applications.

Robust Hybrid Visual Servoing of Omnidirectional Mobile Manipulator With Kinematic Uncertainties Using a Single Camera.

A robust hybrid visual servoing (HVS) method for a single camera-mounted omnidirectional mobile manipulator (OMM) with kinematic uncertainties caused by slipping is proposed herein. Most existing studies pertaining to the visual servoing of mobile manipulators do not consider the kinematic uncertainties and the singularity of the manipulator that can occur during actual operations; furthermore, they required external sensors other than a single camera. In this study, the kinematics of an OMM are modeled considering kinematic uncertainties. Accordingly, an integral sliding-mode observer (ISMO) is designed to estimate the kinematic uncertainties. Subsequently, an integral sliding-mode control (ISMC) law is proposed to achieve robust visual servoing using the estimates of the ISMO. Additionally, an ISMO-ISMC-based HVS method is proposed to address the singularity issue of the manipulator; this method guarantees both robustness and finite-time stability in the presence of kinematic uncertainties. Overall, the entire visual servoing task is performed using only a single camera attached to the end effector without any other external sensors, unlike in previous studies. The stability and performance of the proposed method are verified numerically and experimentally in a slippery environment that generates kinematic uncertainties.

Planar laser-induced incandescence for the study of soot production in a multi-sector RQL Jet A combustor

Understanding the production of non-volatile particulate matter (nvPM), which is composed primarily of soot, is critical not only for reducing emissions but also for improving engine performance. While there has been significant prior work studying the fundamentals of soot formation, there is significantly less work that investigates soot formation with realistic aeroengine geometries, injectors, and fuels in high pressure conditions. In this work, soot production in a three-sector rich-quench-lean (RQL) aeroengine combustor is studied with Jet A fuel. Global equivalence ratios ranging of 0.10 to 0.20 and pressures ranging from 2.7 to 6.9 bar absolute (40 to 100 psia) are tested. In order to characterize in-situ soot production near the fuel injectors, two-dimensional laser-induced incandescence is utilized to estimate single-shot and average soot volume fractions. Time-resolved laser-induced incandescence is then used to create single camera and single laser-shot incandescence decay time images in order to infer how soot particle sizes evolve. Results show a significant increase in soot production at higher global equivalence ratios and higher pressures. Incandescence decay times, however, do not change significantly over the same range of conditions. These measurements can not only help understand soot distributions in practical RQL systems but also help improve future aeroengine combustor designs.

Design of a Multimodal Detection System Tested on Tea Impurity Detection

A multimodal detection system with complementary capabilities for efficient detection was developed for impurity detection. The system consisted of a visible light camera, a multispectral camera, image correction and registration algorithms. It can obtain spectral features and color features at the same time and has higher spatial resolution than a single spectral camera. This system was applied to detect impurities in Pu’er tea to verify its high efficiency. The spectral and color features of each pixel in the images of Pu’er tea were obtained by this system and used for pixel classification. The experimental results showed that the accuracy of a support vector machine (SVM) model based on combined features was 93%, which was 7% higher than that based on spectral features only. By applying a median filtering algorithm and a contour detection algorithm to the label matrix extracted from pixel-classified images, except hair, eight impurities were detected successfully. Moreover, taking advantage of the high resolution of a visible light camera, small impurities could be clearly imaged. By comparing the segmented color image with the pixel-classified image, small impurities such as hair could be detected successfully. Finally, it was proved that the system could obtain multiple images to allow a more detailed and comprehensive understanding of the detected items and had an excellent ability to detect small impurities.

A structural description of pedestrian movement behavior in multiple surveillance videos

This study purpose of recording pedestrians' activities in all surveillance videos and then convert these records into structured descriptions to analyze a large number of surveillance videos. This paper proposes a four-stage processing method. The first stage uses convolutional neural network (CNN) to detect objects in the surveillance video and classify them. The second stage finds the correlation of pedestrians in continuous images and describes multiple objects' trajectories in the surveillance video. The third stage calculates the similarity of the pedestrian object sets of the surveillance video at different times as the basis for pedestrian re-identification. In the fourth stage, extract the walking path of the pedestrian in each surveillance video. Each pedestrian's actions in each surveillance video, including whether to stand, walking path, direction, clothing color and walking time. To construct a correlation table of pedestrian motion in multiple surveillance videos.This experiment uses two image datasets for experimentation and analysis to evaluate this method's effectiveness for pedestrian movement in multiple surveillance videos. The image datasets are the Multi-Camera Object Tracking (MCT) Challenge and PETS 2009 Benchmark Data. This experiment includes pedestrian detection, pedestrian identification, and multi-pedestrian tracking of pedestrians in single and multiple surveillance cameras. The results show that in the pedestrian detection experiment, the convolutional neural network can effectively solve the problems caused by static objects, light changes, similar colors, crowded pedestrians, and incomplete objects. In the pedestrian recognition and re-recognition experiments, the accuracy is reduced due to the influence of low-resolution surveillance videos. This detection method of the pedestrian is more accurate than traditional dynamic object detection methods. After the pedestrian tracking and re-identification are completed, the pedestrian relationships in multiple surveillance cameras are integrated, and a pedestrian behavior association table is built for surveillance videos.

Indirect tuning of a complementary orientation filter using velocity data and a genetic algorithm

In this paper, the accuracy of inertial sensor orientation relative to the level frame is improved through optimal tuning of a complementary filter by a genetic algorithm. While constant filter gains have been used elsewhere, these may introduce errors under dynamic motions when gyroscopes should be trusted more than accelerometers. Optimal gains are prescribed by a Mamdani fuzzy rule base whose membership functions are found using a genetic algorithm and experimental data. Furthermore, model fitness is not based directly on orientation but the error between estimated and ground truth velocities. This paper has three interrelated novel elements. The main novelty is the indirect tuning method, which is simple, low-cost and requires a single camera and inertial sensor. The method is shown to increase tracking accuracy compared with popular baseline filters. Secondary novel elements are the bespoke genetic algorithm and the time agnostic velocity error metric. The contributions from this work can help improve the localization accuracy of assets and human personnel. This research has a direct impact in command and control by improving situational awareness and the ability to direct assets to safe locations using safer routes. This results in increasing safety in applications such as firefighting and battlespace.

Car Parking Occupancy Detection Using Deep Learning

It can be difficult to find a parking space in a busy neighborhood because you never know when one will become available. Due to the rise of automobiles and fall in parking spots, this task even more challenging. To increase the dependability of smart parking systems, numerous researchers have sought to improve them. In smart parking systems, computer vision is a more advantageous method for detecting parking spaces than sensors. Many parking spots may be monitored by a single smart camera, which is also reasonably easy to install and maintain. To identify parking occupancy in camera-captured photos, a CNN model is employed. The CNN model will be trained using the CNRpark+Ext and PKLot datasets, which contain 4081 and 12,416 parking lot photos under various weather scenarios, respectively. The CNN architectures, CmAlexNet and mAlexNet, are five-layer designs built on the eight-layer AlexNet architecture, in which processing speed is increased by removing the completely connected layer and the third and fourth convolutional layers. To increase accuracy, the first convolutional layer of CmAlexNet's filter is resized from 11x11 to 13x13, and batch normalization [1] is used instead of the original method. Three key performance metrics are evaluated between CmAlexNet and mAlexNet: recall, precision, and accuracy in classification. Keywords: Car Parking , Computer Vision , Deep learning , CNN, mAlexnet , cmAlexnet , CNRpark dataset , PKLot dataset.

GaitFormer: Leveraging dual-stream spatial–temporal Vision Transformer via a single low-cost RGB camera for clinical gait analysis

Gait analysis is an essential technique in treating patients with lower limb dysfunctions. Traditional methods often rely on expensive and complex equipment, such as wearable body sensors and a multi-camera with marker tracking system. Aiming for a more cost-effective yet accurate alternative, this paper introduces GaitFormer, a novel approach that leverages Vision Transformer (ViT) for gait analysis using minimal, non-invasive equipment, i.e. a single low-cost RGB camera. Initially, a unique dataset using a multi-camera system with marker tracking, comprising 6 walking patterns gathered from 80 volunteers is developed. The ViT-based GaitFormer is then proposed to automatically recognize human walking patterns through a single RGB camera. GaitFormer comprises hybrid networks for each step, including: (i) a cascaded convolutional 2D human key points estimation network; (ii) a ViT-based dual-stream spatial–temporal network extending the information of human key points into 3D; (iii) leveraging specific lower limb key joints’ angle features for clinical gait analysis, capturing the geometric, kinematic, and physical attributes of human motion; (iv) employing a pure self-attention-based classification network to recognize clinical human walking patterns. The experiments are designed to comprehensively validate each step against various related baseline methods and multi-camera tracking system, with results demonstrating the promising performance of GaitFormer as an affordable, precise, and integrated solution. To the best of our knowledge, GaitFormer is the first hybrid CNN- and ViT-based end-to-end solution via low-cost device for clinically valuable gait analysis.