Search

Motion is vital for life. Currently, the clinical assessment of motion abnormalities is largely qualitative. We previously developed methods to quantitatively assess motion using visual detection systems (around-body) and stretchable electronic sensors (on-body). Here we compare the efficacy of these methods across predefined motions, hypothesizing that the around-body system detects motion with similar accuracy as on-body sensors. Six human volunteers performed six defined motions covering three excursion lengths, small, medium, and large, which were analyzed via both around-body visual marker detection (MoCa version 1.0) and on-body stretchable electronic sensors (BioStamp version 1.0). Data from each system was compared as to the extent of trackability and comparative efficacy between systems. Both systems successfully detected motions, allowing quantitative analysis. Angular displacement between systems had the highest agreement efficiency for the bicep curl and body lean motion, with 73.24% and 65.35%, respectively. The finger pinch motion had an agreement efficiency of 36.71% and chest abduction/adduction had 45.55%. Shoulder abduction/adduction and shoulder flexion/extension motions had the lowest agreement efficiencies with 24.49% and 26.28%, respectively. MoCa was comparable to BioStamp in terms of angular displacement, though velocity and linear speed output could benefit from additional processing. Our findings demonstrate comparable efficacy for non-contact motion detection to that of on-body sensor detection, and offers insight as to the best system selection for specific clinical uses based on the use-case of the desired motion being analyzed. [ABSTRACT FROM AUTHOR]

Fiducial markers play a fundamental role in various fields in which precise localization and tracking are paramount. In Augmented Reality, they provide a known reference point in the physical world so that AR systems can accurately identify, track, and overlay virtual objects. This accuracy is essential for creating a seamless and immersive AR experience, particularly when prompted to cope with the sub-millimeter requirements of medical and industrial applications. This research article presents a comparative analysis of four fiducial marker tracking algorithms, aiming to assess and benchmark their accuracy and precision. The proposed methodology compares the pose estimated by four algorithms running on Hololens 2 with those provided by a highly accurate ground truth system. Each fiducial marker was positioned in 25 sampling points with different distances and orientations. The proposed evaluation method is not influenced by human error, relying only on a high-frequency and accurate motion tracking system as ground truth. This research shows that it is possible to track the fiducial markers with translation and rotation errors as low as 1.36 mm and 0.015 degrees using ArUco and Vuforia, respectively.

There is a gap between mechanical measurement methods under laboratory conditions and those under real conditions of structural monitoring. This paper proposes a method that applies well-established computer vision developments to photomechanics in difficult conditions. It is therefore a technique that is flexible and versatile while maintaining satisfying accuracy. It consists in marker tracking using ArUco fiducial markers as measurement points. It allows locating markers in non-optimal conditions of camera orientation and position. A homography process was used to analyse pictures taken with a view angle. The accuracy of the method was estimated, especially in case of out-of-plane motions, and the impact of the view angle and of the distance between camera and markers on the location error was studied. The method was applied in creep tests to measure crack parameters as well as the transverse expansion of wooden beams. In the application example presented, it enabled to compute distances between markers with only 0.28% of relative error and hence to measure the crack parameters and the long-term shrinkage-swelling of the wooden beams. However, the impacts of brightness variations and camera parameters have not been estimated. This method is very promising when experimental conditions are variable and when multiple measurements are necessary. [ABSTRACT FROM AUTHOR]

Robotic manipulators rely on feedback obtained from rotary encoders for control purposes. This article introduces a vision-based feedback system that can be used in an agricultural context, where the shapes and sizes of fruits are uncertain. We aim to mimic a human, using vision and touch as manipulator control feedback. This work explores the use of a fish-eye lens camera to track a SCARA manipulator with coloured markers on its joints for the position estimation with the goal to reduce costs and increase reliability. The Kalman Filter and the Particle Filter are compared and evaluated in terms of accuracy and tracking abilities of the marker’s positions. The estimated image coordinates of the markers are converted to world coordinates using planar homography, as the SCARA manipulator has co-planar joints and the coloured markers share the same plane. Three laboratory experiments were conducted to evaluate the system’s performance in joint angle estimation of a manipulator. The obtained results are promising, for future cost effective agricultural robotic arms developments. Besides, this work presents solutions and future directions to increase the joint position estimation accuracy.

Nonlinear damping with respect to vibration amplitude is particularly important in mechanical dynamics. The addition of short fibers to damping materials is considered to result in strong nonlinear damping due to interfacial peeling at the edges of the fibers. However, little has been reported on the occurrence of nonlinear damping in short-fiber reinforced rubber due to compounding difficulties. In this study, we investigated the relationship between the damping characteristics and deformation behavior of microdeformed short-fiber reinforced rubber by X-ray computed tomography (CT). We prepared a damping material with a natural rubber (NR) matrix and micrometer-sized polyethylene terephthalate (PET) fiber filler. The loss factor was identified by dynamic mechanical analysis, and three-dimensional strain maps were obtained using marker tracking in the CT data. The addition of 5 wt% PET fibers to NR resulted in an increase in the loss factor. Experimentally, we found that the nonlinear damping of the composite rubber is affected by the peeling of the filler/matrix interface and the strain inside the material. [ABSTRACT FROM AUTHOR]

Featured Application: A potential application of the work is research and design of autonomous underwater vehicle, especially the development of bionicfish-robots. A multi-medium motion capture system based on markers' visual detection is developed and experimentally demonstrated for monitoring underwater intelligent agents such as fish biology and bionic robot-fish. Considering the refraction effect between air and water, a three-dimensional (3D) reconstruction model is established, which can be utilized to reconstruct the 3D coordinate of markers underwater from 2D data. Furthermore, the process of markers matching is undertaken through the multi-lens fusion perception prediction combined K-Means clustering algorithm. Subsequently, in order to track the marker of being occluded, according to the kinematics information of fish, an improved Kalman filtering algorithm is proposed. Finally, the feasibility and effectiveness of proposed system are verified through experimental results. The main models and methods in this paper can provide a reference and inspiration for measurement of underwater intelligent agents. [ABSTRACT FROM AUTHOR]

Background: Fiducial markers are frequently used before treatment for image‐guided patient setup in radiation therapy (RT), but can also be used during treatment for image‐guided intrafraction motion detection. This report describes our implementation of automatic marker detection with periodic kV imaging (TrueBeam v2.5) to monitor and correct intrafraction motion during prostate RT. Methods: We evaluated the reproducibility and accuracy of software fiducial detection using a phantom with 3 implanted fiducial markers. Clinical implementation for patients with intraprostatic fiducials receiving volumetric modulated arc therapy (VMAT) utilized periodic on‐board kV imaging with 10 s intervals during treatment delivery. For each image, the software automatically identified fiducial locations and determined whether their distance relative to planned locations were within a 3 mm tolerance. Motion was corrected if either ≥2 fiducials in a single image or ≥1 fiducial in sequential images were out of tolerance. Results: Phantom studies demonstrated poorer performance of linear fiducials compared to collapsible fiducials, and wide variability to accurately detect fiducials across eight software settings. For any given setting, results were relatively reproducible and precise to ~0.5 mm. Across 17 patients treated with a median of 20 fractions, the software recommended a shift in 44% of fractions, and a shift was actually implemented after visual confirmation of movement greater than the 3 mm threshold in 20% of fractions. Adjustment of our approach led to improved accuracy for the latter (n = 7) patient subset. On average, table repositioning added 3.0 ± 0.3 min to patient time on table. Periodic kV imaging increased skin dose by an estimated 1 cGy per treatment arc. Conclusions: Periodic kV imaging with automatic detection of motion during VMAT prostate treatments is commercially available, and can be successfully implemented to mitigate effects of intrafraction motion with careful attention to software settings. [ABSTRACT FROM AUTHOR]

Fiducial markers are images or landmarks placed in real environment, typically used for pose estimation and camera tracking. Reliable fiducials are strongly desired for many augmented reality (AR) applications, but currently there is no systematic method to design highly reliable fiducials. In this paper, we present fiducial marker optimizer (FMO), a tool to optimize the design attributes of ARToolKit markers, including black to white (B:W) ratio, edge sharpness, and information complexity, and to reduce inter-marker confusion. For these operations, the FMO provides a user friendly interface at the front-end and specialized image processing algorithms at the back-end. We tested manually designed markers and FMO optimized markers in ARToolKit and found that the latter were more robust. The FMO will be used for designing highly reliable fiducials in easy to use fashion. It will improve the application's performance, where it is used.

Human motion capture by optical sensors produces snapshots of the motion of a cloud of points that need to be labeled in order to carry out ensuing motion analysis for medical or other purposes. We generate the labeling of instantaneous captures of the cloud of points, discarding temporal correlations, in the presence of occlusions. Our approach proposes an ensemble of weak classifiers defined over geometrical features extracted from small subsets of the cloud of points. We apply an Adaboost strategy to select a minimal ensemble of weak classifiers achieving a target correct labeling detection accuracy. Furthermore, we use these features to generate the labeling of the points in the cloud even in the presence of occlusions.To deal with the occlusions of markers we search for ensembles of partial labeling solvers which can provide partial consistent labelings which cover the unoccluded markers. We test two greedy search approaches and a genetic algorithm in the search for the optimal ensemble of partial solvers We demonstrate the approach on a real dataset obtained from the measurement of gait motion of persons, with available ground truth labeling. Results are encouraging, achieving high accuracy label generation at a reduced computational cost. [ABSTRACT FROM AUTHOR]

Lower limb injuries, especially those related to the knee joint, are some of the most common and severe injuries among sport practitioners. Consequently, a growing interest in the identification of subjects with high risk of injury has emerged during last years. One of the most commonly used injury risk factor is the measurement of joint angles during the execution of dynamic movements. To that end, techniques such as human motion capture and video analysis have been widely used. However, traditional procedures to measure joint angles present certain limitations, which makes this practice not practical in common clinical settings. This work presents SPIRA, a novel 2D video analysis system directed to support practitioners during the evaluation of joint angles in functional tests. The system employs an infrared camera to track retro-reflective markers attached to the patient's body joints and provide a real-time measurement of the joint angles in a cost-and-time-effective way. The information gathered by the sensor is processed and managed through a computer application that guides the expert during the execution of the tests and expedites the analysis of the results. In order to show the potential of the SPIRA system, a case study has been conducted, performing the analysis with the both the proposed system and a gold-standard in 2D offline video analysis. The results (ICC(ρ ) = 0.996) reveal a good agreement between both tools and prove the reliability of SPIRA. [ABSTRACT FROM AUTHOR]

Innovative lighting and dynamic sound systems as well as adaptive object mapping solutions constitute a rapidly developing branch of lighting technology and multimedia technology. In order to make it possible to adjust the content to specific objects in the scene, it is necessary to correctly identify them and place them in the accepted frame of reference. Dynamic identification and tracking of objects can be carried out based on two particular types of input data: data from markers installed on objects and data from digital recording systems, founding the operation on infrared (IR), visible light (RGB) and the most advanced RGB-D (RGB and depth) analysis. Most systems used today are those that use various types of markers. This paper presents the advantages and disadvantages of such solutions as well as a target system for dynamic identification and mapping of objects and the human body based on the analysis of data from digital RGB-D cameras. Analyses of identification times, implementation of perspective transformations and 3D-to-2D transformations have been carried out in relation to a planar and cuboidal moving surface. Time analyses have been performed in relation to the resolution of registered and processed images.

A multi-medium motion capture system based on markers’ visual detection is developed and experimentally demonstrated for monitoring underwater intelligent agents such as fish biology and bionic robot-fish. Considering the refraction effect between air and water, a three-dimensional (3D) reconstruction model is established, which can be utilized to reconstruct the 3D coordinate of markers underwater from 2D data. Furthermore, the process of markers matching is undertaken through the multi-lens fusion perception prediction combined K-Means clustering algorithm. Subsequently, in order to track the marker of being occluded, according to the kinematics information of fish, an improved Kalman filtering algorithm is proposed. Finally, the feasibility and effectiveness of proposed system are verified through experimental results. The main models and methods in this paper can provide a reference and inspiration for measurement of underwater intelligent agents.

This paper analyzes stability and tracking performance of displayed-marker-based position tracking on visual displays, such as an LCD (Liquid Crystal Display), which have been utilized in several studies about computer-human interaction systems. The tracking system dealt in this paper consists of an LCD and a target object with photo sensors that is placed on the display surface. A fiducial marker image is displayed beneath the object and the photo sensors detect the relative displacement between the marker and the object. The detected displacement is fed back to the tracking program and the program updates the marker position such that it will track the object. In general, this kind of tracking methods can suffer from display latency that arises from programs, graphic engines, and internal signal processing circuits of LCDs. This paper investigates the latency characteristics to reveal that the lag is not constant; it fluctuates with time. The paper, then, formulates the tracking system and analyzes how the lag affects the tracking stability. Then, the paper analyzes the tracking performance of two different classes of stable tracking algorithms, which are PD control and lag compensation. Based on the analyses, the paper provides a guideline on the selection of tracking algorithms, as well as tracking parameters. The analyzed results, as well as the guideline, are verified by experiments in 1-DOF horizontal motions.

Objective. The prostate moves in accordance with the movement of surrounding organs. Tumor position can change by ≥3 mm during radiotherapy. Given the difficulties of visualizing the prostate fluoroscopically, fiducial markers are generally implanted into the prostate to monitor its motion during treatment. Recently, internally motion guidance methods of the prostate using a 99.5% gold/0.5% iron flexible notched wire fiducial marker (Gold Anchor® , Naslund Medical AB, Huddinge, Sweden), which requires a 22 gauge needle, has been used. However, because the notched wire can retain its linear shape, acquire a spiral shape, or roll into an irregular ball, detecting it on fluoroscopic images in real-time incurs higher computation costs. Approach. We developed a fiducial tracking algorithm to achieve real-time computation. The marker is detected on the first image frame using a shape filter that employs inter-class variance for the marker likelihood calculated by the filter, focusing on the large difference in densities between the marker and its surroundings. After the second frame, the marker is tracked by adding to the shape filter the similarity to the template cropped from the area around the marker position detected in the first frame. We retrospectively evaluated the algorithm's marker tracking accuracy for ten prostate cases, analyzing two fractions in each case. Main results. Tracking positional accuracy averaged over all patients was 0.13 ± 0.04 mm (mean ± standard deviation, Euclidean distance) and 0.25 ± 0.09 mm (95th percentile). Computation time was 2.82 ± 0.20 ms/frame averaged over all frames. Significance. Our algorithm successfully and stably tracked irregularly-shaped markers in real time., (© 2024 Institute of Physics and Engineering in Medicine.)

Growth and development of winter crops is strongly limited by low temperature during winter. Monitoring the temporal dynamics and thermal limits of leaf growth in that period can give important insights into the growth physiology at low temperature, crop management and future breeding traits for winter crops. In this study, we focussed on winter rapeseed as a model, dicotyledonous winter crop to study leaf growth under natural winter field conditions. Leaf growth was measured using a high-resolution marker based image sequence analysis method and the results were evaluated in the context of environmental conditions. Leaves stopped growing at a base temperature of 0°C. Above ~4°C, leaves grew with a diel (24 h) growth rhythm, which is typically known for dicots at thermally non-limiting growth conditions. Relative leaf growth rates at temperatures above this 4°C threshold were higher at night and showed a pronounced depression during the day, which we could describe by a model based on the environmental factors vapour pressure deficit (VPD), temperature and light with VPD exerting the strongest negative effect on leaf growth. We conclude that leaf growth of the selected model species at low temperatures shows a transition between pronounced environmental regulation and a superposition of environmental and internal, possibly circadian-clock-dependent regulation. Winter crops such as rapeseed are exposed to thermal limiting growth conditions for long periods of their development. Monitoring the temporal dynamics and thermal limits of leaf growth under winter field conditions helps in understanding growth physiology at low temperature and to adopt breeding strategies for future climate scenarios. Rapeseed leaves show a daily growth rhythm with higher rates at night and limited growth during the day regulated by environmental as well as endogenous circadian factors. [ABSTRACT FROM AUTHOR]

Background and objective To develop and validate a robust and cost-effective 3D respiratory monitoring system based on a Kinect device with a custom-made simple marker. Methods A 3D respiratory monitoring system comprising the simple marker and the Microsoft Kinect v2 device was developed. The marker was designed for simple and robust detection, and the tracking algorithm was developed using the depth, RGB, and infra-red images acquired from the Kinect sensor. A Kalman filter was used to suppress movement noises. The major movements of the marker attached to the four different locations of body surface were determined from the initially collected tracking points of the marker while breathing. The signal level of respiratory motion with the tracking point was estimated along the major direction vector. The accuracy of the results was evaluated through a comparison with those of the conventional stereovision navigation system (NDI Polaris Spectra). Results Sixteen normal volunteers were enrolled to evaluate the accuracy of this system. The correlation coefficients between the respiratory motion signal from the Kinect device and conventional navigation system ranged from 0.970 to 0.999 and from 0.837 to 0.995 at the abdominal and thoracic surfaces, respectively. The respiratory motion signal from this system was obtained at 27–30 frames/s. Conclusions This system with the Kinect v2 device and simple marker could be used for cost-effective, robust and accurate 3D respiratory motion monitoring. In addition, this system is as reliable for respiratory motion signal generation and as practically useful as the conventional stereovision navigation system and is less sensitive to patient posture. [ABSTRACT FROM AUTHOR]

In many industrial applications, the quality of mixing between different materials is fundamental to guarantee the desired properties of products. However, properly modeling and understanding polymer mixing presents noticeable difficulties, because of the variety and complexity of the phenomena involved. This is also the case with the Cavity Transfer Mixer (CTM), an add‐on to be mounted downstream of existing extruders, to improve distributive mixing. The present work proposes a fully three‐dimensional model of the CTM: a finite element solver provides the transient velocity field, which is used in the mapping method implementation to compute the concentration field evolution and quantify mixing. Several simulations are run assessing the impact on mixing of geometrical and functioning parameters. In general, the number of cavities per row should be limited and the cavity size rather big to guarantee good mixing quality. © 2017 American Institute of Chemical Engineers AIChE J, 64: 1034–1048, 2018 [ABSTRACT FROM AUTHOR]

The optical tracking information in manufacturing can provide valuable support and time saving for autonomous operations, but ill environment conditions prevent a better performance of vision systems. In this work, a method for estimating object position under semi-controlled environment where lighting conditions change dynamically is proposed. This method incorporates regression analysis that combines light measurement and an augmented reality (AR) system. Augmented reality (AR) combines virtual objects with real environment. Furthermore, every AR application uses a video camera to capture an image including a marker in order to place a virtual object, which gives user an enriched environment. Using a tracking system to estimate the marker's position with respect to the camera coordinate frame is needed to positioning a virtual object. Most research studies on tracking system for AR are under controlled environment. The problem is that tracking systems for markers are sensitive to variations in lighting conditions in the real environment. To solve this problem, a method is proposed to better estimate a marker position based on regression analysis, where lighting conditions are taken into account. Our approach improves the accuracy of the marker position estimation under different lighting conditions. The experimental data obtained under a laboratory context with changes on light condition are fitted with this approach with an accuracy of 99 %. [ABSTRACT FROM AUTHOR]

The article presents a hybrid monitoring technique for the measurement of the deformation field. The goal is to obtain information about crack propagation in existing structures, for the purpose of monitoring their state of health. The measurement technique is based on the capture and analysis of a digital image set. Special markers were used on the surface of the structures that can be removed without damaging existing structures as the historical masonry. The digital image analysis was done using software specifically designed in Matlab to follow the tracking of the markers and determine the evolution of the deformation state. The method can be used in any type of structure but is particularly suitable when it is necessary not to damage the surface of structures. A series of experiments carried out on masonry walls of the Oliverian Museum (Pesaro, Italy) and Palazzo Silvi (Perugia, Italy) have allowed the validation of the procedure elaborated by comparing the results with those derived from traditional measuring techniques. [ABSTRACT FROM AUTHOR]

We propose a novel hybrid inertial sensors-based indoor pedestrian dead reckoning system, aided by computer vision-derived position measurements. In contrast to prior vision-based or vision-aided solutions, where environmental markers were used—either deployed in known positions or extracted directly from it—we use a shoe-fixed marker, which serves as positional reference to an opposite shoe-mounted camera during foot swing, making our system self-contained. Position measurements can be therefore more reliably fed to a complementary unscented Kalman filter, enhancing the accuracy of the estimated travelled path for 78%, compared to using solely zero velocities as pseudo-measurements.

Marker-based XROMM requires software tools for: (1) correcting fluoroscope distortion; (2) calibrating X-ray cameras; (3) tracking radio-opaque markers; and (4) calculating rigid body motion. In this paper we describe and validate XMALab, a new open-source software package for marker-based XROMM (C++ source and compiled versions on Bitbucket). Most marker-based XROMM studies to date have used XrayProject in MATLAB. XrayProject can produce results with excellent accuracy and precision, but it is somewhat cumbersome to use and requires a MATLAB license. We have designed XMALab to accelerate the XROMM process and to make it more accessible to new users. Features include the four XROMMsteps (listed above) in one cohesive user interface, real-time plot windows for detecting errors, and integration with an online data management system, XMAPortal. Accuracy and precision of XMALab when tracking markers in a machined object are ±0.010 and ±0.043 mm, respectively. Mean precision for nine users tracking markers in a tutorial dataset of minipig feeding was ±0.062 mm in XMALab and ±0.14 mm in XrayProject. Reproducibility of 3D point locations across nine users was 10-fold greater in XMALab than in XrayProject, and six degree-of-freedom bone motions calculated with a joint coordinate system were 3- to 6-fold more reproducible in XMALab. XMALab is also suitable for tracking white or black markers in standard light videos with optional checkerboard calibration. We expect XMALab to increase both the quality and quantity of animal motion data available for comparative biomechanics research. [ABSTRACT FROM AUTHOR]

Leaf growth in monocot crops such as wheat and barley largely follows the daily temperature course, particularly under cold but humid springtime field conditions. Knowledge of the temperature response of leaf extension, particularly variations close to the thermal limit of growth, helps define physiological growth constraints and breedingrelated genotypic differences among cultivars. Here, we present a novel method, called 'Leaf Length Tracker' (LLT), suitable for measuring leaf elongation rates (LERs) of cereals and other grasses with high precision and high temporal resolution under field conditions. The method is based on image sequence analysis, using a marker tracking approach to calculate LERs. We applied the LLT to several varieties of winter wheat (Triticum aestivum), summer barley (Hordeum vulgare), and ryegrass (Lolium perenne), grown in the field and in growth cabinets under controlled conditions. LLT is easy to use and we demonstrate its reliability and precision under changing weather conditions that include temperature, wind, and rain. We found that leaf growth stopped at a base temperature of 0°C for all studied species and we detected significant genotype-specific differences in LER with rising temperature. The data obtained were statistically robust and were reproducible in the tested environments. Using LLT, we were able to detect subtle differences (sub-millimeter) in leaf growth patterns. This method will allow the collection of leaf growth data in a wide range of future field experiments on different graminoid species or varieties under varying environmental or treatment conditions. [ABSTRACT FROM AUTHOR]

The Rubber Cord Adhesion Inflation Test (RCAIT) has been recently introduced to characterize the bond quality between coated steel wires and rubber as used in tires applications. In this experiment, a pressurized fluid is used to initiate and propagate a delamination along the reinforcement. A constant pressure is measured during the crack propagation regime so that using adequate energy balance analysis, the critical Strain Energy Release Rate (SERR) can be determined. Due to the axisymmetric nature of the specimen, the crack tip is not visible nor other information which would ease the analysis of the fracture process. Then, new experimental developments and experimental data reduction techniques are introduced to demonstrate the reliability of the RCAIT. [ABSTRACT FROM AUTHOR]

Objective . Using MV images for real-time image guided radiation therapy (IGRT) is ideal as it does not require additional imaging equipment, adds no additional imaging dose and provides motion data in the treatment beam frame of reference. However, accurate tracking using MV images is challenging due to low contrast and modulated fields. Here, a novel real-time marker tracking system based on a convolutional neural network (CNN) classifier was developed and evaluated on retrospectively acquired patient data for MV-based IGRT for prostate cancer patients. Approach . MV images, acquired from 29 volumetric modulated arc therapy (VMAT) prostate cancer patients treated in a multi-institutional clinical trial, were used to train and evaluate a CNN-based marker tracking system. The CNN was trained using labelled MV images from 9 prostate cancer patients (35 fractions) with implanted markers. CNN performance was evaluated on an independent cohort of unseen MV images from 20 patients (78 fractions), using a Precision-Recall curve (PRC), area under the PRC plot (AUC) and sensitivity and specificity. The accuracy of the tracking system was evaluated on the same unseen dataset and quantified by calculating mean absolute (±1 SD) and [1st, 99th] percentiles of the geometric tracking error in treatment beam co-ordinates using manual identification as the ground truth. Main results . The CNN had an AUC of 0.99, sensitivity of 98.31% and specificity of 99.87%. The mean absolute geometric tracking error was 0.30 ± 0.27 and 0.35 ± 0.31 mm in the lateral and superior-inferior directions of the MV images, respectively. The [1st, 99th] percentiles of the error were [-1.03, 0.90] and [-1.12, 1.12] mm in the lateral and SI directions, respectively. Significance . The high classification performance on unseen MV images demonstrates the CNN can successfully identify implanted prostate markers. Furthermore, the sub-millimetre accuracy and precision of the marker tracking system demonstrates potential for adaptation to real-time applications., (© 2023 Institute of Physics and Engineering in Medicine.)