This paper proposes an algorithm that extracts features of back side of the vehicle and detects the front vehicle in real-time by local feature tracking of vehicle in the continuous images.The features in back side of...This paper proposes an algorithm that extracts features of back side of the vehicle and detects the front vehicle in real-time by local feature tracking of vehicle in the continuous images.The features in back side of the vehicle are vertical and horizontal edges,shadow and symmetry.By comparing local features using the fixed window size,the features in the continuous images are tracked.A robust and fast Haarlike mask is used for detecting vertical and horizontal edges,and shadow is extracted by histogram equalization,and the sliding window method is used to compare both side templates of the detected candidates for extracting symmetry.The features for tracking are vertical edges,and histogram is used to compare location of the peak and magnitude of the edges.The method using local feature tracking in the continuous images is more robust for detecting vehicle than the method using single image,and the proposed algorithm is evaluated by continuous images obtained on the expressway and downtown.And it can be performed on real-time through applying it to the embedded system.展开更多
This paper proposed a robust method based on the definition of Mahalanobis distance to track ground moving target. The feature and the geometry of airborne ground moving target tracking systems are studied at first. B...This paper proposed a robust method based on the definition of Mahalanobis distance to track ground moving target. The feature and the geometry of airborne ground moving target tracking systems are studied at first. Based on this feature, the assignment relation of time-nearby target is calculated via Mahalanobis distance, and then the corresponding transformation formula is deduced. The simulation results show the correctness and effectiveness of the proposed method.展开更多
Real-time seam tracking can improve welding quality and enhance welding efficiency during the welding process in automobile manufacturing.However,the teaching-playing welding process,an off-line seam tracking method,i...Real-time seam tracking can improve welding quality and enhance welding efficiency during the welding process in automobile manufacturing.However,the teaching-playing welding process,an off-line seam tracking method,is still dominant in automobile industry,which is less flexible when welding objects or situation change.A novel real-time algorithm consisting of seam detection and generation is proposed to track seam.Using captured 3D points,space vectors were created between two adjacent points along each laser line and then a vector angle based algorithm was developed to detect target points on the seam.Least square method was used to fit target points to a welding trajectory for seam tracking.Furthermore,the real-time seam tracking process was simulated in MATLAB/Simulink.The trend of joint angles vs.time was logged and a comparison between the off-line and the proposed seam tracking algorithm was conducted.Results show that the proposed real-time seam tracking algorithm can work in a real-time scenario and have high accuracy in welding point positioning.展开更多
A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects...A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm.This article examines methods and tools for recognizing and tracking the class of small moving objects,such as ants.To fulfill those aims,a customized You Only Look Once Ants Recognition(YOLO_AR)Convolutional Neural Network(CNN)has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool.The proposed model is an extension of the You Only Look Once v4(Yolov4)512×512 model with an additional Self Regularized Non–Monotonic(Mish)activation function.Additionally,the scalable solution for continuous object recognizing and tracking was implemented.This solution is based on the OpenDatacam system,with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line.During the study,the methods of the alignment algorithm for finding the trajectory of moving objects were modified.I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree(k-d tree)matching algorithm used in OpenDataCam.Remarkably,such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives(FP).Therefore,I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking(MOT)benchmark.Furthermore,additional customization parameters for object recognition and tracking results parsing and filtering were added,like boundary angle threshold(BAT)and past frames trajectory prediction(PFTP).Experimental tests confirmed the results of the study on a mobile device.During the experiment,parameters such as the quality of recognition and tracking of moving objects,the PFTP and BAT,and the configuration parameters of the neural network and boundary line model were analyzed.The results showed an increased tracking accuracy with the proposed methods by 50%.The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.展开更多
Radio-frequency(RF) tomography is an emerging technology which derives targets location information by analyzing the changes of received signal strength(RSS) in wireless links. This paper presents and evaluates a nove...Radio-frequency(RF) tomography is an emerging technology which derives targets location information by analyzing the changes of received signal strength(RSS) in wireless links. This paper presents and evaluates a novel RF tomography system which is capable of detecting and tracking a time-varying number of targets in a cluttered indoor environment. The system incorporates an observation model based on RSS attenuation histogram and a multi-target tracking-by-detection filtering approach based on probability hypothesis density(PHD) filter. In addition, the sequential Monte Carlo method is applied to implement the multi-target filtering. To evaluate the tracking system, the experiments involving up to 3 targets were performed within an obstructed indoor area of 70 m2. The experimental results indicate that the proposed tracking system is capable of tracking a time-varying number of targets.展开更多
This paper proposes a block Mean-Shift algorithm based on target real-time update and LBP texture features, through the target update improves the accuracy of target tracking, enhances the local character of the targe...This paper proposes a block Mean-Shift algorithm based on target real-time update and LBP texture features, through the target update improves the accuracy of target tracking, enhances the local character of the target through the target block, so as to improve the robustness of algorithm based on skin color backgrounds. And then analyze the Mean-Shift algorithm cannot recover quickly lost target tracking defects, and its improvement by combining the frame difference method.展开更多
Vehicle tracking plays a crucial role in intelligent transportation, autonomous driving, and video surveillance. However, challenges such as occlusion, multi-target interference, and nonlinear motion in dynamic scenar...Vehicle tracking plays a crucial role in intelligent transportation, autonomous driving, and video surveillance. However, challenges such as occlusion, multi-target interference, and nonlinear motion in dynamic scenarios make tracking accuracy and stability a focus of ongoing research. This paper proposes an integrated method combining YOLOv8 object detection with adaptive Kalman filtering. The approach employs a support vector machine (SVM) to dynamically select the optimal filter (including standard Kalman filter, extended Kalman filter, and unscented Kalman filter), enhancing the system’s adaptability to different motion patterns. Additionally, an error feedback mechanism is incorporated to dynamically adjust filter parameters, further improving responsiveness to sudden events. Experimental results on the KITTI and UA-DETRAC datasets demonstrate that the proposed method significantly improves detection accuracy (mAP@0.5 increased by approximately 3%), tracking accuracy (MOTA improved by 5%), and system robustness, providing an efficient solution for vehicle tracking in complex environments.展开更多
Probability Hypothesis Density(PHD)filtering approach has shown its advantages in tracking time varying number of targets even when there are noise,clutter and misdetection.For linear Gaussian Mixture(GM)system,PHD fi...Probability Hypothesis Density(PHD)filtering approach has shown its advantages in tracking time varying number of targets even when there are noise,clutter and misdetection.For linear Gaussian Mixture(GM)system,PHD filter has a closed form recursion(GMPHD).But PHD filter cannot estimate the trajectories of multi-target because it only provides identity-free estimate of target states.Existing data association methods still remain a big challenge mostly because they are com-putationally expensive.In this paper,we proposed a new data association algorithm using GMPHD filter,which significantly alleviated the heavy computing load and performed multi-target trajectory tracking effectively in the meantime.展开更多
Dynamically tracking hundreds of individual pits is essential to determine whether there exist "hot spots" for the formation of clathrin-coated pits or if the pits formed randomly on the plasma membrane. We ...Dynamically tracking hundreds of individual pits is essential to determine whether there exist "hot spots" for the formation of clathrin-coated pits or if the pits formed randomly on the plasma membrane. We propose an automated approach to detect these particles based on an improved á trous wavelet transform decomposition with automatic threshold selection and post processing solution, and to track the dynamic process with a greedy algorithm. The results indicate that the detection method can successfully detect most particles in an image with accuracy of 98.61% and 97.65% for adaptor and clathrin images, respectively, and that the tracking algorithm can resolve merging and splitting issues encountered when analyzing dynamic, live-cell images of clathrin assemblies.展开更多
In traffic scenarios,the dynamic characteristics and random behaviors of vehicles are the main reasons for the frequent occurrence of collision accidents.Traditional early warning systems restrict traffic safety due t...In traffic scenarios,the dynamic characteristics and random behaviors of vehicles are the main reasons for the frequent occurrence of collision accidents.Traditional early warning systems restrict traffic safety due to low detection accuracy and poor tracking effect.Research has proposed a vehicle safety distance early warning system based on deep learning to enhance traffic safety.Innovations include:adopting the self-calibrated illumination(SCI)algorithm to overcome light interference;the YOLOv11 algorithm is improved by introducing a secondary innovative cross-domain feature attention(CDFA)mechanism,reconstructing the feature pyramid,and integrating knowledge distillation to balance detection accuracy and real-time performance.The DeepSORT algorithm is improved by applying group convolution to reduce the number of parameters and replacing Intersection over Union(IoU)with MPDIoU to enhance tracking accuracy.The distance between vehicles is calculated by using the monocular vision ranging method.The detection,tracking,and ranging modules are integrated into a vehicle safety distance early warning system.Experimental evaluation demonstrates a marked improvement in the system's performance.On the public dataset,the detection model exhibits a gain of 3.89%in mAP@0.5 and 2.76%in mAP@0.5:0.95,while the tracking model achieves a 0.9%increase in multiple object tracking accuracy(MOTA).Furthermore,real-world vehicle validation confirms that the synergistic operation of the detection and tracking modules effectively mitigates the miss rate,thereby substantiating a tangible enhancement in overall system safety.展开更多
基金supported by the Brain Korea 21 Project in 2011 and MKE(The Ministry of Knowledge Economy),Korea,under the ITRC(Infor mation Technology Research Center)support program supervised by the NIPA(National IT Industry Promotion Agency)(NIPA-2011-C1090-1121-0010)
文摘This paper proposes an algorithm that extracts features of back side of the vehicle and detects the front vehicle in real-time by local feature tracking of vehicle in the continuous images.The features in back side of the vehicle are vertical and horizontal edges,shadow and symmetry.By comparing local features using the fixed window size,the features in the continuous images are tracked.A robust and fast Haarlike mask is used for detecting vertical and horizontal edges,and shadow is extracted by histogram equalization,and the sliding window method is used to compare both side templates of the detected candidates for extracting symmetry.The features for tracking are vertical edges,and histogram is used to compare location of the peak and magnitude of the edges.The method using local feature tracking in the continuous images is more robust for detecting vehicle than the method using single image,and the proposed algorithm is evaluated by continuous images obtained on the expressway and downtown.And it can be performed on real-time through applying it to the embedded system.
基金Supported by the National Natural Science Foundation of China Youth Science Fund Project(Nos.62101405,61372185)
文摘This paper proposed a robust method based on the definition of Mahalanobis distance to track ground moving target. The feature and the geometry of airborne ground moving target tracking systems are studied at first. Based on this feature, the assignment relation of time-nearby target is calculated via Mahalanobis distance, and then the corresponding transformation formula is deduced. The simulation results show the correctness and effectiveness of the proposed method.
基金Supported by Ministerial Level Advanced Research Foundation(65822576)Beijing Municipal Education Commission(KM201310858004,KM201310858001)
文摘Real-time seam tracking can improve welding quality and enhance welding efficiency during the welding process in automobile manufacturing.However,the teaching-playing welding process,an off-line seam tracking method,is still dominant in automobile industry,which is less flexible when welding objects or situation change.A novel real-time algorithm consisting of seam detection and generation is proposed to track seam.Using captured 3D points,space vectors were created between two adjacent points along each laser line and then a vector angle based algorithm was developed to detect target points on the seam.Least square method was used to fit target points to a welding trajectory for seam tracking.Furthermore,the real-time seam tracking process was simulated in MATLAB/Simulink.The trend of joint angles vs.time was logged and a comparison between the off-line and the proposed seam tracking algorithm was conducted.Results show that the proposed real-time seam tracking algorithm can work in a real-time scenario and have high accuracy in welding point positioning.
文摘A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm.This article examines methods and tools for recognizing and tracking the class of small moving objects,such as ants.To fulfill those aims,a customized You Only Look Once Ants Recognition(YOLO_AR)Convolutional Neural Network(CNN)has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool.The proposed model is an extension of the You Only Look Once v4(Yolov4)512×512 model with an additional Self Regularized Non–Monotonic(Mish)activation function.Additionally,the scalable solution for continuous object recognizing and tracking was implemented.This solution is based on the OpenDatacam system,with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line.During the study,the methods of the alignment algorithm for finding the trajectory of moving objects were modified.I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree(k-d tree)matching algorithm used in OpenDataCam.Remarkably,such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives(FP).Therefore,I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking(MOT)benchmark.Furthermore,additional customization parameters for object recognition and tracking results parsing and filtering were added,like boundary angle threshold(BAT)and past frames trajectory prediction(PFTP).Experimental tests confirmed the results of the study on a mobile device.During the experiment,parameters such as the quality of recognition and tracking of moving objects,the PFTP and BAT,and the configuration parameters of the neural network and boundary line model were analyzed.The results showed an increased tracking accuracy with the proposed methods by 50%.The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.
文摘Radio-frequency(RF) tomography is an emerging technology which derives targets location information by analyzing the changes of received signal strength(RSS) in wireless links. This paper presents and evaluates a novel RF tomography system which is capable of detecting and tracking a time-varying number of targets in a cluttered indoor environment. The system incorporates an observation model based on RSS attenuation histogram and a multi-target tracking-by-detection filtering approach based on probability hypothesis density(PHD) filter. In addition, the sequential Monte Carlo method is applied to implement the multi-target filtering. To evaluate the tracking system, the experiments involving up to 3 targets were performed within an obstructed indoor area of 70 m2. The experimental results indicate that the proposed tracking system is capable of tracking a time-varying number of targets.
文摘This paper proposes a block Mean-Shift algorithm based on target real-time update and LBP texture features, through the target update improves the accuracy of target tracking, enhances the local character of the target through the target block, so as to improve the robustness of algorithm based on skin color backgrounds. And then analyze the Mean-Shift algorithm cannot recover quickly lost target tracking defects, and its improvement by combining the frame difference method.
文摘Vehicle tracking plays a crucial role in intelligent transportation, autonomous driving, and video surveillance. However, challenges such as occlusion, multi-target interference, and nonlinear motion in dynamic scenarios make tracking accuracy and stability a focus of ongoing research. This paper proposes an integrated method combining YOLOv8 object detection with adaptive Kalman filtering. The approach employs a support vector machine (SVM) to dynamically select the optimal filter (including standard Kalman filter, extended Kalman filter, and unscented Kalman filter), enhancing the system’s adaptability to different motion patterns. Additionally, an error feedback mechanism is incorporated to dynamically adjust filter parameters, further improving responsiveness to sudden events. Experimental results on the KITTI and UA-DETRAC datasets demonstrate that the proposed method significantly improves detection accuracy (mAP@0.5 increased by approximately 3%), tracking accuracy (MOTA improved by 5%), and system robustness, providing an efficient solution for vehicle tracking in complex environments.
基金Supported by the National Natural Science Foundation of China(No.60772154)the President Foundation of Graduate University of Chinese Academy of Sciences(No.085102GN00)
文摘Probability Hypothesis Density(PHD)filtering approach has shown its advantages in tracking time varying number of targets even when there are noise,clutter and misdetection.For linear Gaussian Mixture(GM)system,PHD filter has a closed form recursion(GMPHD).But PHD filter cannot estimate the trajectories of multi-target because it only provides identity-free estimate of target states.Existing data association methods still remain a big challenge mostly because they are com-putationally expensive.In this paper,we proposed a new data association algorithm using GMPHD filter,which significantly alleviated the heavy computing load and performed multi-target trajectory tracking effectively in the meantime.
基金supported by the National Basic Research Program of China (2011CB707900)the National Natural Science Foundation of China (10974093, 11174141)the Fundamental Research Funds for the Central Universities (1103020402, 1116020410 and 1112020401)
文摘Dynamically tracking hundreds of individual pits is essential to determine whether there exist "hot spots" for the formation of clathrin-coated pits or if the pits formed randomly on the plasma membrane. We propose an automated approach to detect these particles based on an improved á trous wavelet transform decomposition with automatic threshold selection and post processing solution, and to track the dynamic process with a greedy algorithm. The results indicate that the detection method can successfully detect most particles in an image with accuracy of 98.61% and 97.65% for adaptor and clathrin images, respectively, and that the tracking algorithm can resolve merging and splitting issues encountered when analyzing dynamic, live-cell images of clathrin assemblies.
基金supported in part by the National Natural Science Foundation of China(No.52172389).
文摘In traffic scenarios,the dynamic characteristics and random behaviors of vehicles are the main reasons for the frequent occurrence of collision accidents.Traditional early warning systems restrict traffic safety due to low detection accuracy and poor tracking effect.Research has proposed a vehicle safety distance early warning system based on deep learning to enhance traffic safety.Innovations include:adopting the self-calibrated illumination(SCI)algorithm to overcome light interference;the YOLOv11 algorithm is improved by introducing a secondary innovative cross-domain feature attention(CDFA)mechanism,reconstructing the feature pyramid,and integrating knowledge distillation to balance detection accuracy and real-time performance.The DeepSORT algorithm is improved by applying group convolution to reduce the number of parameters and replacing Intersection over Union(IoU)with MPDIoU to enhance tracking accuracy.The distance between vehicles is calculated by using the monocular vision ranging method.The detection,tracking,and ranging modules are integrated into a vehicle safety distance early warning system.Experimental evaluation demonstrates a marked improvement in the system's performance.On the public dataset,the detection model exhibits a gain of 3.89%in mAP@0.5 and 2.76%in mAP@0.5:0.95,while the tracking model achieves a 0.9%increase in multiple object tracking accuracy(MOTA).Furthermore,real-world vehicle validation confirms that the synergistic operation of the detection and tracking modules effectively mitigates the miss rate,thereby substantiating a tangible enhancement in overall system safety.
