Track Loss Research Articles

Toward a navigation framework for fetoscopy

PurposeFetoscopic laser photocoagulation of placental anastomoses is the most effective treatment for twin-to-twin transfusion syndrome (TTTS). A robust mosaic of placenta and its vascular network could support surgeons’ exploration of the placenta by enlarging the fetoscope field-of-view. In this work, we propose a learning-based framework for field-of-view expansion from intra-operative video frames.MethodsWhile current state of the art for fetoscopic mosaicking builds upon the registration of anatomical landmarks which may not always be visible, our framework relies on learning-based features and keypoints, as well as robust transformer-based image-feature matching, without requiring any anatomical priors. We further address the problem of occlusion recovery and frame relocalization, relying on the computed features and their descriptors.ResultsExperiments were conducted on 10 in-vivo TTTS videos from two different fetal surgery centers. The proposed framework was compared with several state-of-the-art approaches, achieving higher textrm{SSIM}_{5} on 7 out of 10 videos and a success rate of 93.25% in occlusion recovery.ConclusionThis work introduces a learning-based framework for placental mosaicking with occlusion recovery from intra-operative videos using a keypoint-based strategy and features. The proposed framework can compute the placental panorama and recover even in case of camera tracking loss where other methods fail. The results suggest that the proposed framework has large potential to pave the way to creating a surgical navigation system for TTTS by providing robust field-of-view expansion.

Joint Power and Bandwidth Allocation with RCS Fluctuation Characteristic for Space Target Tracking

Reasonable allocation of space-based radar resources is a crucial aspect of improving the accuracy of space multi-target tracking and enhancing spatial awareness. The conventional resource allocation algorithm fails to exploit the high dynamic radar cross-section (RCS) characteristics, resulting in poor tracking robustness, tracking divergence, or even loss of tracking. However, the RCS of space targets fluctuates considerably in actual tracking scenarios, which cannot be disregarded for space target tracking tasks. To address this issue, we propose an adaptive allocation method that considers the dynamic RCS fluctuation characteristic for space-based radar tracking assignments. The proposed method exploits the predictable orbital information of space target to calculate the real-time observation angle of radar, and then obtains the multi-target dynamic RCS through the target RCS dataset. By combining the obtained RCS sequence, radar power, and bandwidth, an optimal model for radar tracking accuracy is established based on the multi-target posterior Cramér–Rao lower bound (PCRLB) to evaluate the tracking performance. By resolving the aforementioned multivariance optimization problem, we eventually acquire the results of power and bandwidth pre-allocation for tracking multiple space targets. Simulation results validate that, compared with the traditional methods, the proposed joint dynamic RCS power and bandwidth allocation (JRPBA) method can achieve superior tracking accuracy and minimize instances of missed tracking.

A Re-detection Target Tracking Framework after Loss for Unmanned Aerial Vehicle

Currently, the target tracking algorithm based on discriminative correlation filter (DCF) and deep learning (DL) plays an increasingly important role in UAV (unmanned aerial vehicle). However, existing algorithms have limitations such as limited search region, difficulty in re-capturing targets when tracking is lost, and high computational complexity, which makes them difficult to apply to the UAV platform. In this paper, a target re-detection tracking algorithm (RDT) after tracking loss is designed for mobile platforms. RDT uses DCF in simple scenes to track targets and establish a target model between different frames. To ensure the algorithm when tracking is lost due to occlusion or moving out of view, an efficient switching criterion is designed to indirectly invoke the detection algorithm to re-capture the target and guide online learning of DCF. And the experiment results on the OTB benchmark show that RDT can re-capture the target after tracking is lost, and the speed of operation on the CPU is 85fps.

Evaluation and improvement of HMD-based and RGB-based hand tracking solutions in VR

Hand tracking has become a state-of-the-art technology in the modern generation of consumer VR devices. However, off-the-shelf solutions do not support hand detection for more than two hands at the same time at distances beyond arm’s length. The possibility to track multiple hands at larger distances would be beneficial for colocated multi-user VR scenarios, allowing user-worn devices to track the hands of other users and therefore reducing motion artifacts caused by hand tracking loss. With the global focus of enabling natural hand interactions in colocated multi-user VR, we propose an RGB image input-based hand tracking method, built upon the MediaPipe framework, that can track multiple hands at once at distances of up to 3 m. We compared our method’s accuracy to that of Oculus Quest and Leap Motion, at different distances from the tracking device and in static and dynamic settings. The results of our evaluation show that our method provides only slightly less accurate results than Oculus Quest or Leap motion in the near range (with median errors below 1.75 cm at distances below 75 cm); at larger distances, its accuracy remains stable (with a median error of 4.7 cm at the distance of 2.75 m) while Leap Motion and Oculus Quest either loose tracking or produce very inaccurate results. Taking into account the broad choice of suitable hardware (any RGB camera) and the ease of setup, our method can be directly applied to colocated multi-user VR scenarios.

Single-Point Least Square Matching Embedded Method for Improving Visual SLAM

Visual simultaneous location and mapping (VSLAM) has been widely used today, but for some tough cases with insufficient feature points (such as weak texture and motion blur), it is still an ongoing challenge to further boost the corresponding robustness and accuracy. In this paper, to address the problem of inadequate feature points, for conventional visual SLAM, we improve the original feature matching mechanism by providing more correspondences with higher position precision. Firstly, based on the original feature matching results in the map initialization stage, a single-point least square matching (LSM) which iteratively refines position of the matching points by considering geometric and photometric consistency, is presented to improve the map initialization result. Secondly, to improve the tracking performance, given the motion model, the single-point LSM is again used to yield new correspondences from the unsuccessfully matched features of original matching mechanism between current frame and previous frame in the tracking thread, and these extra correspondences participate in and boost the subsequent pose optimization. Lastly, for newly inserted keyframe, we focus on identifying the corresponding 3D map points’ reprojections on relevant local neighbor keyframes, these reprojections are added in the local bundle adjustment (BA) optimization for further improving the estimation of camera pose and local map. The popular VSLAM package, ORB-SLAM2, is used to demonstrate the efficacy of the proposed improvements. An ablation study on accuracy and tracking stability on public datasets are reported, furthermore, the comparisons with several state of the arts (namely, VINS-Mono, ORB-SLAM3, DSO, PL-SLAM) are investigated. The experimental results show that our approach can improve accuracy in weak texture scenes and motion blur cases, reduce the number of tracking loss and improve tracking robustness.

An Improved Pig Counting Algorithm Based on YOLOv5 and DeepSORT Model.

Pig counting is an important task in pig sales and breeding supervision. Currently, manual counting is low-efficiency and high-cost and presents challenges in terms of statistical analysis. In response to the difficulties faced in pig part feature detection, the loss of tracking due to rapid movement, and the large counting deviation in pig video tracking and counting research, this paper proposes an improved pig counting algorithm (Mobile Pig Counting Algorithm with YOLOv5xpig and DeepSORTPig (MPC-YD)) based on YOLOv5 + DeepSORT model. The algorithm improves the detection rate of pig body parts by adding two different sizes of SPP networks and using SoftPool instead of MaxPool operations in YOLOv5x. In addition, the algorithm includes a pig reidentification network, a pig-tracking method based on spatial state correction, and a pig counting method based on frame number judgment on the DeepSORT algorithm to improve pig tracking accuracy. Experimental analysis shows that the MPC-YD algorithm achieves an average precision of 99.24% in pig object detection and an accuracy of 85.32% in multitarget pig tracking. In the aisle environment of the slaughterhouse, the MPC-YD algorithm achieves a correlation coefficient (R2) of 98.14% in pig counting from video, and it achieves stable pig counting in a breeding environment. The algorithm has a wide range of application prospects.

Target tracking algorithm based on occlusion prediction

Tracking failure caused by occlusion is one of the most challenging problems in visual tracking tasks. The current Siamese network-based object tracking algorithms use the maximum confidence strategy and center cropping mechanism, which may lead to partial or complete loss of target tracking. To address this problem, we propose an occlusion prediction-based tracking method, which uses an occlusion prediction branch to evaluate the occlusion degree of the target in four directions: up, down, left and right, and combines information fusion to predict the target position, and uses dynamic background suppression to reduce the interference of similar targets. First, we use dynamic background suppression to preprocess the input image, weaken the interference around the target, then use the target location information and regression information output by the Siamese network to obtain the candidate target position, and finally use the occlusion prediction information for information fusion to obtain the final target position. Although our method brings additional computational burden, it can still achieve real-time performance. Experiments show that our proposed method can effectively improve the robustness of Siamese tracking algorithms under different occlusion conditions. In the interference test based on OTB100, our method improves the tracking success rate by about 7% and the precision by about 10% for SiamBan and SiamGat algorithms respectively; in the interference test based on VOT2018, our method can effectively reduce the tracking loss frequency of Siamban and SiamGat algorithms, improving the EAO(Expected Average Overlap) by about 10%.

An Improved Unscented Kalman Filtering Combined with Feature Triangle for Head Position Tracking

Aiming at the problem of feature point tracking loss caused by large head rotation and facial occlusion in doctors, this paper designs a head-position-tracking system based on geometric triangles and unscented Kalman filtering. By interconnecting the three feature points of the left and right pupil centers and the tip of the nose, they form a coplanar triangle. When the posture of the doctor’s head changes due to rotation, the shape of the corresponding geometric triangle will also deform. Using the inherent laws therein, the head posture can be estimated based on changes in the geometric model. Due to the inaccurate positioning of feature points caused by the deflection of the human head wearing a mask, traditional linear Kalman filtering algorithms are difficult to accurately track feature points. This paper combines geometric triangles with an unscented Kalman Filter (UKF) to obtain head posture, which has been fully tested in different environments, such as different faces, wearing/not wearing masks, and dark/bright light via public and measured datasets. The final experimental results show that compared to the linear Kalman filtering algorithm with a single feature point, the traceless Kalman filtering algorithm combined with geometric triangles in this paper not only improves the robustness of nonlinear angle of view tracking but also can provide more accurate estimates than traditional Kalman filters.

Quantification of motion during microvascular anastomosis simulation using machine learning hand detection.

Microanastomosis is one of the most technically demanding and important microsurgical skills for a neurosurgeon. A hand motion detector based on machine learning tracking technology was developed and implemented for performance assessment during microvascular anastomosis simulation. A microanastomosis motion detector was developed using a machine learning model capable of tracking 21 hand landmarks without physical sensors attached to a surgeon's hands. Anastomosis procedures were simulated using synthetic vessels, and hand motion was recorded with a microscope and external camera. Time series analysis was performed to quantify the economy, amplitude, and flow of motion using data science algorithms. Six operators with various levels of technical expertise (2 experts, 2 intermediates, and 2 novices) were compared. The detector recorded a mean (SD) of 27.6 (1.8) measurements per landmark per second with a 10% mean loss of tracking for both hands. During 600 seconds of simulation, the 4 nonexperts performed 26 bites in total, with a combined excess of motion of 14.3 (15.5) seconds per bite, whereas the 2 experts performed 33 bites (18 and 15 bites) with a mean (SD) combined excess of motion of 2.8 (2.3) seconds per bite for the dominant hand. In 180 seconds, the experts performed 13 bites, with mean (SD) latencies of 22.2 (4.4) and 23.4 (10.1) seconds, whereas the 2 intermediate operators performed a total of 9 bites with mean (SD) latencies of 31.5 (7.1) and 34.4 (22.1) seconds per bite. A hand motion detector based on machine learning technology allows the identification of gross and fine movements performed during microanastomosis. Economy, amplitude, and flow of motion were measured using time series data analysis. Technical expertise could be inferred from such quantitative performance analysis.

Vision-Based Real-Time Motion Estimation of Ground Vehicles from Aerial Imagery

This research presents vision-based maneuvering object motion estimation in case of occlusion. Unequal dimension Interactive Multiple Model (UDIMM) approach is applied to increase the motion prediction accuracy when no measurement is available. Current deep learning-based multi-object tracking algorithms cannot track the objects when an occlusion exists, even for quite a short time. The main reason is the motion model used in tracking algorithms. If the measurement coming from the object detector is not available, the prediction model propagates the Kalman Filter’s motion model. In this period, the object may maneuver or accelerate/decelerate so that the tracker cannot assign the tracklets successfully. Eventually, the ID switch occurs, and object tracking gets lost. Using the UDIMM approach, track loss due to the ID switch is highly reduced compared to the simple constant velocity motion model used in the StrongSort multi-object tracker algorithm. A two-dimensional simulation environment is created to compare the ground truth data with the estimation results. Performance results of different filters are compared.

Rate-induced tipping: thresholds, edge states and connecting orbits

Rate-induced tipping (R-tipping) occurs when time-variation of input parameters of a dynamical system interacts with system timescales to give genuine nonautonomous instabilities. Such instabilities appear as the input varies at some critical rates and cannot, in general, be understood in terms of autonomous bifurcations in the frozen system with a fixed-in-time input. This paper develops an accessible mathematical framework for R-tipping in multidimensional nonautonomous dynamical systems with an autonomous future limit. We focus on R-tipping via loss of tracking of base attractors that are equilibria in the frozen system, due to crossing what we call regular R-tipping thresholds. These thresholds are anchored at infinity by regular R-tipping edge states: compact normally hyperbolic invariant sets of the autonomous future limit system that have one unstable direction, orientable stable manifold, and lie on a basin boundary. We define R-tipping and critical rates for the nonautonomous system in terms of special solutions that limit to a compact invariant set of the autonomous future limit system that is not an attractor. We focus on the case when the limit set is a regular edge state, introduce the concept of edge tails, and rigorously classify R-tipping into reversible, irreversible, and degenerate cases. The central idea is to use the autonomous dynamics of the future limit system to analyse R-tipping in the nonautonomous system. We compactify the original nonautonomous system to include the limiting autonomous dynamics. Considering regular R-tipping edge states that are equilibria allows us to prove two results. First, we give sufficient conditions for the occurrence of R-tipping in terms of easily testable properties of the frozen system and input variation. Second, we give necessary and sufficient conditions for the occurrence of reversible and irreversible R-tipping in terms of computationally verifiable (heteroclinic) connections to regular R-tipping edge states in the autonomous compactified system.

Automatic tracking method for submarine cables and pipelines of AUV based on side scan sonar

Submarine cables and pipelines laid on the seafloor are critical infrastructures for energy transportation and communication transmission and must be inspected in a timely manner to determine maintenance needs. Autonomous inspection of cables (including pipelines) and their surroundings using underwater vehicles is difficult because of sensor myopia and deviations in cable routing. This study proposes an automatic cable-tracking method (ACTM) based on side-scan sonar (SSS) that endows autonomous underwater vehicles (AUVs) with autonomous decision-making capabilities to achieve fast localization and stable tracking of submarine cables with large uncertainties. Compared with the traditional predefined waypoint method, the proposed method decomposes the tasks into a series of independent behaviors according to operational objectives; thus, the AUV can perform online dynamic task replanning and realize an autonomous coverage search of the target region. When a cable is detected, the cable-tracking task is constructed as a path-following control problem on the horizontal plane. To reduce uncertainties caused by environmental noise and sensor measurement errors, a uniform motion model based on target tracking theory was used to model the virtual mass abstracted from the cable, and the states of the cable were estimated using the Kalman filter. Subsequently, an adaptive line-of-sight guidance algorithm for the SSS is established to dynamically adjust the look-ahead distance by introducing states such as cable curvature and angular change rate, which somewhat alleviates the tracking loss caused by the SSS short-sightedness. The proposed ACTM effectiveness and robustness were validated through numerical simulations and field tests. The results showed that the proposed method can achieve automatic search and adaptive tracking of cables, further reducing the role of humans in the task.

Research and implementation of non-linear management and monitoring system for classified information network

Abstract In order to better realize the secret-related information monitoring system, an algorithm based on a nonlinear network is proposed and is combined with the traditional algorithm. This article mainly analyzes the theory of nonlinear networks, designs and trains new network parameters according to their own needs, and combines the nonlinear network as a feature extractor with the existing intrusion detection and wandering detection algorithms, which greatly improves the recognition ability of traditional algorithms. The main feature of a nonlinear network is that it can extract the positional features of objects from the network while also extracting object features, that is, positioning and classification are realized in the same network. As a feature extractor, this network can not only have a higher recognition rate than background difference, hog, and other algorithms but also have a greater ability to extract position information than other convolutional neural networks. The successful application of nonlinear production network systems in TV stations at all levels has greatly improved the editing and production capability and efficiency of TV programs. How to ensure the safe, reliable, stable, orderly, and efficient operation of nonlinear production network systems requires vendors and TVS Taiwan technical staff to jointly conduct in-depth research and summarize their findings. In this article, from the perspective of TV users, information components in nonlinear production network systems are analyzed, including class, title management mode, storage space management, material management, security management, and workflow management in nonlinear systems. Make some analysis, discussion, and summaries of network system and operation management problems. The experimental results show that the nonlinear algorithm in this article has a significant advantage over the original tracking algorithm; that is, most tracking algorithms do not have the ability of category recognition during the initial tracking process, which means that these tracking algorithms cannot accurately know what they are tracking. Because the nonlinear network has the ability to output categories, whether it is initial tracking or tracking loss recovery, nonlinearity has fundamentally better advantages than other tracking algorithms. Therefore, it can be predicted that there is a strong recognition ability in the later monitoring and wandering detection. It has been proved that the nonlinear algorithm can be effectively applied to the secret information monitoring system.

‘The fact that I know I can do it is quite a motivator now’: a qualitative study exploring experiences maintaining weight loss 6 months after completing a weight loss programme for knee osteoarthritis

ObjectiveTo explore experiences maintaining weight loss 6 months after completing a multicomponent weight loss programme for knee osteoarthritis.DesignQualitative study based on an interpretivist paradigm and a phenomenological approach that was...

Semantic closed-loop based visual mapping algorithm for automated valet parking

Simultaneous localization and mapping (SLAM) is one of the core technologies to realize automated valet parking (AVP). Currently, advanced visual feature-based SLAM systems suffer from feature extraction failure and tracking loss due to the constraints of textureless scenes, unclear illumination, and dynamic conditions. To address these problems, this paper proposes a visual SLAM algorithm based on a semantic closed-loop detection algorithm using surround-view cameras and inertial measurement units (IMU) as sensors. The algorithm combines semantic features and the idea of inverse index to improve the traditional keyframes selection methods and the closed-loop detection algorithms, effectively avoiding the tedious and complicated feature point matching and improving the computational efficiency of the computer. Experiments show that the algorithm in this paper achieves better results in terms of precision and recall, absolute trajectory error (ATE), and relative pose error (RPE), and can meet the demand for SLAM and subsequent navigation in indoor parking lots.

Convolutional Neural Network-Based Automated System for Dog Tracking and Emotion Recognition in Video Surveillance

This paper proposes a multi–convolutional neural network (CNN)-based system for the detection, tracking, and recognition of the emotions of dogs in surveillance videos. This system detects dogs in each frame of a video, tracks the dogs in the video, and recognizes the dogs’ emotions. The system uses a YOLOv3 model for dog detection. The dogs are tracked in real time with a deep association metric model (DeepDogTrack), which uses a Kalman filter combined with a CNN for processing. Thereafter, the dogs’ emotional behaviors are categorized into three types—angry (or aggressive), happy (or excited), and neutral (or general) behaviors—on the basis of manual judgments made by veterinary experts and custom dog breeders. The system extracts sub-images from videos of dogs, determines whether the images are sufficient to recognize the dogs’ emotions, and uses the long short-term deep features of dog memory networks model (LDFDMN) to identify the dog’s emotions. The dog detection experiments were conducted using two image datasets to verify the model’s effectiveness, and the detection accuracy rates were 97.59% and 94.62%, respectively. Detection errors occurred when the dog’s facial features were obscured, when the dog was of a special breed, when the dog’s body was covered, or when the dog region was incomplete. The dog-tracking experiments were conducted using three video datasets, each containing one or more dogs. The highest tracking accuracy rate (93.02%) was achieved when only one dog was in the video, and the highest tracking rate achieved for a video containing multiple dogs was 86.45%. Tracking errors occurred when the region covered by a dog’s body increased as the dog entered or left the screen, resulting in tracking loss. The dog emotion recognition experiments were conducted using two video datasets. The emotion recognition accuracy rates were 81.73% and 76.02%, respectively. Recognition errors occurred when the background of the image was removed, resulting in the dog region being unclear and the incorrect emotion being recognized. Of the three emotions, anger was the most prominently represented; therefore, the recognition rates for angry emotions were higher than those for happy or neutral emotions. Emotion recognition errors occurred when the dog’s movements were too subtle or too fast, the image was blurred, the shooting angle was suboptimal, or the video resolution was too low. Nevertheless, the current experiments revealed that the proposed system can correctly recognize the emotions of dogs in videos. The accuracy of the proposed system can be dramatically increased by using more images and videos for training the detection, tracking, and emotional recognition models. The system can then be applied in real-world situations to assist in the early identification of dogs that may exhibit aggressive behavior.

PLDS-SLAM: Point and Line Features SLAM in Dynamic Environment

Visual simultaneous localization and mapping (SLAM), based on point features, achieves high localization accuracy and map construction. They primarily perform simultaneous localization and mapping based on static features. Despite their efficiency and high precision, they are prone to instability and even failure in complex environments. In a dynamic environment, it is easy to keep track of failures and even failures in work. The dynamic object elimination method, based on semantic segmentation, often recognizes dynamic objects and static objects without distinction. If there are many semantic segmentation objects or the distribution of segmentation objects is uneven in the camera view, this may result in feature offset and deficiency for map matching and motion tracking, which will lead to problems, such as reduced system accuracy, tracking failure, and track loss. To address these issues, we propose a novel point-line SLAM system based on dynamic environments. The method we propose obtains the prior dynamic region features by detecting and segmenting the dynamic region. It realizes the separation of dynamic and static objects by proposing a geometric constraint method for matching line segments, combined with the epipolar constraint method of feature points. Additionally, a dynamic feature tracking method based on Bayesian theory is proposed to eliminate the dynamic noise of points and lines and improve the robustness and accuracy of the SLAM system. We have performed extensive experiments on the KITTI and HPatches datasets to verify these claims. The experimental results show that our proposed method has excellent performance in dynamic and complex scenes.

Research and application of downhole drilling depth based on computer vision technique

Gas drainage and water detection and release by drilling are essential to prevent gas and water disasters in China. The accurate drilling depth substantially affects gas drainage, water exploration, and safety problems. Consequently, it is essential to precisely monitor drilling depth by counting drill pipe to assure drilling efficacy. Based on increasingly sophisticated subterranean video surveillance technology, this paper leverages computer vision technology to implement precise drill pipe counting and calculates the corresponding drilling depth determined by the number of drill pipes. First, the operation of the drilling rig was examined, and the movement trajectory of the impact power head on the drilling rig feeding mechanism was estimated to quantify the number of drilling pipes. The optimized Retinex mine image enhancement algorithm is used to preprocess the initial underground image with issues including an unreliable light source, considerable light and shadow fluctuation, and blurred lens. Furthermore, the challenges of fixed scale and easy loss of target tracking in the Kernel Correlation Filter Algorithm are addressed by employing an adaptive scale algorithm and a re-detection approach to boost target location data accuracy. Eventually, the amount of drill pipes is calculated by the effective peak count of the displacement, the precise drilling depth is achieved, and the result is consistent with the actual drilling construction condition by evaluating the on-site drilling video. This study adopted computer vision tracking technology to tackle the challenge of calculating drilling depth and building progress accurately, ensuring the safe and efficient output of the mine.

Tracking Multiple Unmanned Aerial Vehicles through Occlusion in Low-Altitude Airspace

In an intelligent multi-target tracking (MTT) system, the tracking filter cannot track multi-targets significantly through occlusion in a low-altitude airspace. The most challenging issues are the target deformation, target occlusion and targets being concealed by the presence of background clutter. Thus, the true tracks that follow the desired targets are often lost due to the occlusion of uncertain measurements detected by a sensor, such as a motion capture (mocap) sensor. In addition, sensor measurement noise, process noise and clutter measurements degrade the system performance. To avoid track loss, we use the Markov-chain-two (MC2) model that allows the propagation of target existence through the occlusion region. We utilized the MC2 model in linear multi-target tracking based on the integrated probabilistic data association (LMIPDA) and proposed a modified integrated algorithm referred to here as LMIPDA-MC2. We consider a three-dimensional surveillance for tracking occluded targets, such as unmanned aerial vehicles (UAVs) and other autonomous vehicles at low altitude in clutters. We compared the results of the proposed method with existing Markov-chain model based algorithms using Monte Carlo simulations and practical experiments. We also provide track retention and false-track discrimination (FTD) statistics to explain the significance of the LMIPDA-MC2 algorithm.

An Effective Motion-Tracking Scheme for Machine-Learning Applications in Noisy Videos

Detecting and tracking objects of interest in videos is a technology that can be used in various applications. For example, identifying cell movements or mutations through videos obtained in real time can be useful information for decision making in the medical field. However, depending on the situation, the quality of the video may be below the expected level, and in this case, it may be difficult to check necessary information. To overcome this problem, we proposed a technique to effectively track objects by modifying the simplest color balance (SCB) technique. An optimal object detection method was devised by mixing the modified SCB algorithm and a binarization technique. We presented a method of displaying object labels on a per-frame basis to track object movements in a video. Detecting objects and tagging labels through this method can be used to generate object motion-based prediction training data for machine learning. That is, based on the generated training data, it is possible to implement an artificial intelligence model for an expert system based on various object motion measurements. As a result, the main object detection accuracy in noisy videos was more than 95%. This method also reduced the tracking loss rate to less than 10%.