Center Location Error Research Articles

Zynq FPGA-Based Acceleration of Kernelized Correlation Filters via High-Level Synthesis of a Custom DFT Block

This study presents a hardware-software co-design implementation of an accelerator for the Kernelized Correlation Filter (KCF) tracking algorithm. Leveraging the High-level synthesis (HLS) and the Zynq heterogeneous platform, the KCF algorithm’s performance is enhanced by using a custom hardware implementation for the computationally intensive Discrete Fourier Transform (DFT) operation. Within this framework, a custom combined DFT and inverse DFT IP, named CDFT, is developed and optimized on the Programmable Logic (PL) side of the Xilinx ZCU102 FPGA, whereas the rest of the KCF algorithm is run with customized Petalinux build on the (Processing System) side. To assess real-world performance, a driver for the CDFT IP and a user application were created to measure metrics like Center Location Error (CLE), Intersection over Union (IoU), and Frame per Second (FPS). The designed DFT accelerator achieves a remarkable speedup of 21x compared to a software DFT implementation. At the algorithm level, the KCF accelerator obtains a 6x speed up with negligible precision loss. In comparison to prior studies employing exclusively hardware implementations, the proposed approach demonstrates a high accuracy at a moderate speed, while there exists potential for further optimizations to enhance its performance even further.

An effective object detection and tracking using automated image annotation with inception based faster R-CNN model

The present study advances object detection and tracking techniques by proposing a novel model combining Automated Image Annotation with Inception v2-based Faster RCNN (AIA-IFRCNN). The research methodology utilizes the DCF-CSRT model for image annotation, Faster RCNN for object detection, and the inception v2 model for feature extraction, followed by a softmax layer for image classification. The proposed AIA-IFRCNN model is evaluated on three benchmark datasets: Bird (Dataset 1), UCSDped2 (Dataset 2), and Under Water (Dataset 3), to determine prediction accuracy, annotation time, Center Location Error (CLE), and Overlap Rate (OR). The experimental results indicate that the AIA-IFRCNN model outperformed existing models regarding detection accuracy and tracking performance. Notably, it achieved a maximum detection accuracy of 95.62 % on Dataset 1, outperforming other models. Additionally, it achieved minimum average CLE values of 4.16, 5.78, and 3.54, and higher overlap rates of 0.92, 0.90, and 0.94 on the respective datasets (1, 2 and 3). Hence, this research work on object detection and tracking using the AIA-IFRCNN model is essential for improving system efficiency and fostering innovation in the field of computer vision and object tracking.

Deep features fusion for KCF-based moving object tracking

Real-time object tracking and occlusion handling are critical research areas in computer vision and machine learning. Developing an efficient and accurate object-tracking method that can operate in real-time while handling occlusion is essential for various applications, including surveillance, autonomous driving, and robotics. However, relying solely on a single hand-crafted feature results in less robust tracking. As a hand-crafted feature extraction technique, HOG effectively detects edges and contours, which is essential in localizing objects in images. However, it does not capture fine details in object appearance and is sensitive to changes in lighting conditions. On the other hand, the grayscale feature has computational efficiency and robustness to changes in lighting conditions. The deep feature can extract features that express the image in more detail and discriminate between different objects. By fusing different features, the tracking method can overcome the limitations of individual features and capture a complete representation of the object. The deep features can be generated with transfer learning networks. However, selecting the right network is difficult, even in real-time applications. This study integrated the deep feature architecture and hand-crafted features HOG and grayscale in the KCF method to solve this problem. The object images were obtained through at least three convolution blocks of transfer learning architecture, such as Xception, DenseNet, VGG16, and MobileNet. Once the deep feature was extracted, the HOG and grayscale features were computed and combined into a single stack. In the KCF method, the stacked features acquired the actual object location by conveying a maximum response. The result shows that this proposed method, especially in the combination of Xception, grayscale, and HOG features, can be implemented in real-time applications with a small center location error.

A Variational Method for Analyzing Vortex Flows in Radar-Scanned Tornadic Mesocyclones. Part IV: Applications to the 20 May 2013 Oklahoma Tornadic Mesocyclone

Abstract The variational method for vortex flow (VF) analyses, called VF-Var (formulated in Part I), is applied to the 20 May 2013 Newcastle–Moore tornadic mesocyclone observed from the operational KTLX radar and an experimental phased-array radar. The dual-Doppler-analyzed VF field reveals the following features: The axisymmetric part of the VF is a well-defined slantwise two-cell vortex in which the maximum tangential velocity is nearly 40 m s−1 at the edge of the vortex core (0.6 km from the vortex center), the central downdraft velocity reaches −35 m s−1 at 3-km height, and the surrounding-updraft velocity reaches 26 m s−1 at 5-km height. The total VF field is a loosely defined slantwise two-cell vortex consisting of a nearly axisymmetric vortex core (in which the ground-relative surface wind speed reaches 50 m s−1 on the southeast edge), a strong nonaxisymmetric slantwise downdraft in the vortex core, and a main updraft in a banana-shaped area southeast of the vortex core, which extends slantwise upward and spirals cyclonically around the vortex core. The single-Doppler analysis with observations from the KTLX radar only exhibits roughly the same features as the dual-Doppler analysis but contains spurious vertical-motion structures in and around the vortex core. Analysis errors are assessed by leveraging the findings from Parts II and III, which indicate that the dual-Doppler-analyzed VF is accurate enough to represent the true VF but the single-Doppler-analyzed VF is not (especially for nonaxisymmetric vertical motions in and around the vortex core), so the dual-Doppler-analyzed VF should be useful for initializing/verifying high-resolution tornado simulations. Significance Statement After the variational method for vortex flow (VF) analyses, called VF-Var (formulated in Part I of this paper series), was tested successfully with simulated radar observations in Part II and its sensitivity to vortex center location error was examined in Part III, the method is now applied to the 20 May 2013 Newcastle–Moore tornadic mesocyclone observed from the operational KTLX radar and an experimental phased-array radar. Analysis errors are assessed by leveraging the findings from Parts II and III. The results indicate that the dual-Doppler-analyzed VF is accurate enough to represent the true VF (although the single-Doppler-analyzed VF is not especially for nonaxisymmetric vertical motions in and around the vortex core) and thus should be useful for initializing/verifying high-resolution tornado simulations.

A Variational Method for Analyzing Vortex Flows in Radar-Scanned Tornadic Mesocyclones. Part III: Sensitivities to Vortex Center Location Errors

Abstract When the vortex center location is estimated from a radar-scanned tornadic mesocyclone, the estimated location is not error-free. This raises an important issue concerning the sensitivities of analyzed vortex flow (VF) fields by the VF-Var (formulated in Part I of this paper series and tested in Part II) to vortex center location errors, denoted by Δxc. Numerical experiments are performed to address this issue with the following findings: The increase of |Δxc| from zero to a half of vortex core radius causes large analysis error increases in the vortex core but the increased analysis errors decrease rapidly away from the vortex core especially for dual-Doppler analyses. The increased horizontal-velocity errors in the vortex core are mainly in the Δxc-normal component, because this component varies much more rapidly than the other component along the Δxc direction in the vortex core. The vertical variations of Δxc distort the vertical correlation structure of Δxc-dislocated VF-dependent background error covariance, which can increase the analysis errors in the vortex core. The dual-Doppler analyses have adequate accuracies outside the vortex core even when |Δxc| increases to a half of vortex core radius, while single-Doppler analyses can also have adequate accuracies outside the vortex core mainly for the single-Doppler-observed velocity component. The sensitivities to Δxc are largely unaffected by the vortex slanting. The above findings are important and useful for assessing the accuracies of analyzed VFs for real radar-observed tornadic mesocyclones. Significance Statement When the vortex center location is estimated from a radar-scanned tornadic mesocyclone, the estimated location is not error-free. This raises an issue concerning the sensitivity of analyzed vortex flow (VF) by the VF-Var (formulated in Part I of this paper series and tested with simulated radar observations in Part II) to vortex center location error. This issue and its required investigations are very important for the VF-Var to be applied to real radar-observed tornadic mesocyclones, especially in an operational setting with the WSR-88Ds. Numerical experiments are performed to address this issue. The findings from these experiments are important and useful for assessing the accuracies of VF-Var analyzed VF fields for real radar-observed tornadic mesocyclones.

Improved Multidomain Convolutional Neural Network for Intelligent Tracking of Marangoni Effect Process Sequence Images

Sequence image target tracking has two requirements of accuracy and speed. Aiming at the problem of slow speed of convolutional neural network (CNN) target tracking algorithm, an improved algorithm for sequence image target tracking is designed by using a fusion of Faster multidomain CNN (MDNet) and optical flow method. The target algorithm designed in this article is improved on the basis of MDNet. The optical flow method is used to obtain the primary selection box first, which can effectively deal with the challenges brought by scale change and in-plane rotation, and has good tracking speed and performance. The objective algorithm of this article is used to explore the Marangoni effect in the melting process of solid iron tailings in molten blast furnace slag. Because the main component of iron tailings is Silica, and the melting point of Silica is the highest, so taking Silica as the research object, an intelligent analysis of Marangoni effect in high-temperature melting process of Silica particles is realized. The results show that the target algorithm achieves the real-time standard (FPS > 24), the average distance precision (DP) is 97.3%, the average success rate (SR) is 98.02%, the average center location error (CLE) is 6.2684 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">μ</i> m, and the average overlap precision (OP) coverage is 63.86%. The target algorithm has certain advantages in the following five evaluation indexes: DP, SR, CLE, OP, and frames per second.

Multiple Cues-Based Robust Visual Object Tracking Method

Visual object tracking is still considered a challenging task in computer vision research society. The object of interest undergoes significant appearance changes because of illumination variation, deformation, motion blur, background clutter, and occlusion. Kernelized correlation filter- (KCF) based tracking schemes have shown good performance in recent years. The accuracy and robustness of these trackers can be further enhanced by incorporating multiple cues from the response map. Response map computation is the complementary step in KCF-based tracking schemes, and it contains a bundle of information. The majority of the tracking methods based on KCF estimate the target location by fetching a single cue-like peak correlation value from the response map. This paper proposes to mine the response map in-depth to fetch multiple cues about the target model. Furthermore, a new criterion based on the hybridization of multiple cues i.e., average peak correlation energy (APCE) and confidence of squared response map (CSRM), is presented to enhance the tracking efficiency. We update the following tracking modules based on hybridized criterion: (i) occlusion detection, (ii) adaptive learning rate adjustment, (iii) drift handling using adaptive learning rate, (iv) handling, and (v) scale estimation. We integrate all these modules to propose a new tracking scheme. The proposed tracker is evaluated on challenging videos selected from three standard datasets, i.e., OTB-50, OTB-100, and TC-128. A comparison of the proposed tracking scheme with other state-of-the-art methods is also presented in this paper. Our method improved considerably by achieving a center location error of 16.06, distance precision of 0.889, and overlap success rate of 0.824.

Real-Time FEA-based breast deformation simulation using artificial neural network

Background and objectiveTreatment of breast cancer involves two stages: diagnosis and treatment. It is difficult to correlate the imaging results at the two stages because as the patient’s position changes during treatment, the images captured during diagnosis do not represent the tumor location during the treatment. In the absence of real-time imaging during treatment, the visualization of tumor location is challenging for surgeons. In this paper, we combined finite element analysis (FEA) and neural network (NN) to accurately simulation real-time breast deformation for different patient’s positions. MethodsThis research focuses on investigating visualization of breast deformation for different patients positions. We collected magnetic resonance imaging (MRI) images of a patient during diagnosis. This data was preprocessed to form a 3D reconstructed model that was used to run a FEA simulation. FEA simulates the deformation of breast tissues for different constraints, such as glandular ratio and gravity angle. However, it typically takes FEA to simulation such deformation from as fast as a few minutes to around 40 minutes when 8 cores were used. To obtain real-time visualization, we constructed a NN model that takes a breast gravity angle and glandular / fat ratio (breast material) as input to estimate breast deformation for different positions offline. The NN predicts the deformation and helps to visualize breast deformation in real-time (5 ms prediction time). ResultsOur results show that NN has the ability to map the gravity direction and breast material structure to the breast shape and tumor location accurately, while, keeping run time to a minimum. The average nodal displacement validation error of breast surface was 0.013 mm, and the tumor center location error was 0.05 mm. ConclusionsWe took a patient’s diagnostic images as input, and used FEA to simulate breast deformation. Then we implemented two NNs, one to take gravity direction and material composition as input to predict breast surface nodal displacement; the other one used the same input, but output tumor center position. Finally, the model was used to provide a real-time breast deformation prediction during treatment (5ms). Our approach can be used to create tools that can be used inside an operating room for accurate tumor prediction.

Visual tracking using multi-layer appearance approach

This paper investigated single target tracking of arbitrary objects. Tracking is a difficult problem due to a variety of challenges such as scale variation, motion, background clutter, illumination etc. To achieve better tracking performance under these severe conditions, this paper proposed covariance descriptor based on multi-layer instance search region. Our results show that the proposed approach significantly improves the performance in term of centre location error (in pixels) compared to covariance descriptor with using a fixed bounding box. From this work, it is believed that we have constructed a great solution in choosing best layer for this descriptor. This will be addressed in the next future work such as consider target motion during tracking.

A Comparison of HWRF Six-Hourly 4DEnVar and Hourly 3DEnVar Assimilation of Inner Core Tail Dopper Radar Observations for the Prediction of Hurricane Edouard (2014)

Six-hourly three-dimensional ensemble variational (3DEnVar) (6H-3DEnVar) data assimilation (DA) assumes constant background error covariance (BEC) during a six-hour DA window and is, therefore, unable to account for temporal evolution of the BEC. This study evaluates the one-hourly 3DEnVar (1H-3DEnVar) and six-hourly 4DEnVar (6H-4DEnVar) DA methods for the analyses and forecasts of hurricanes with rapidly evolving BEC. Both methods account for evolving BEC in a hybrid EnVar DA system. In order to compare these methods, experiments are conducted by assimilating inner core Tail Doppler Radar (TDR) wind for Hurricane Edouard (2014) and by running the Hurricane Weather Research and Forecasting (HWRF) model. In most metrics, 1H-3DEnVar and 6H-4DEnVar analyses and forecasts verify better than 6H-3DEnVar. 6H-4DEnVar produces better thermodynamic analyses than 1H-3DEnVar. Radar reflectivity shows that 1H-3DEnVar produces better structure forecasts. For the first 24–48 h of the intensity forecast, 6H-4DEnVar forecast performs better than 1H-3DEnVar verified against the best track. Degraded 1H-3DEnVar forecasts are found to be associated with background storm center location error as a result of underdispersive ensemble storm center spread. Removing location error in the background improves intensity forecasts of 1H-3DEnVar.

New performance measures for object tracking under complex environments

Various performance measures based on the ground truth and without ground truth exist to evaluate the quality of a developed tracking algorithm. The existing popular measures—average center location error (acle) and average tracking accuracy (ata) based on ground truth—may sometimes create confusion to quantify the quality of a developed algorithm for tracking an object under some complex environments (e.g., scaled or oriented or both scaled and oriented object). In this article, we propose three new auxiliary performance measures based on ground truth information to evaluate the quality of a developed tracking algorithm under such complex environments. Moreover, one performance measure is developed by combining both two existing measures (acle and ata) and three new proposed measures for better quantifying the developed tracking algorithm under such complex conditions. Some examples and experimental results conclude that the proposed measure is better than existing measures to quantify one developed algorithm for tracking objects under such complex environments.

PUPIL VISUAL TRACKING ALGORITHMS FOR AUTOMATED STATIC PERIMETRY SYSTEMS

Abstract. Some diseases, for instance, a glaucoma, cause visual field defects. For the timely diagnostics of such defects, various methods are used. One of the state-of-the-art diagnostic methods is automated static perimetry. The method of static perimetry consists in the light sensitivity determination in different parts of the visual field using stationary objects of variable luminosity. When scanning the visual field in this way, an important factor is the control of gaze fixation at the fixation point. The greatest accuracy in determining the gaze fixation position is achieved by the method of the pupil visual tracking using a video camera.In this paper, four groups of visual tracking algorithms are considered: segmentation-based methods, correlation methods, methods based on optical flow and on weighted average. An experimental comparison of these methods was carried out using the base of video recordings obtained in the automatic static perimetry apparatus. On these videos the ground truth tracks of pupil were marked. The comparison was conducted according to two criteria: center location error and tracking length. It is shown that only the weighted average method has an acceptable tracking length.

Evolutionary Rear-Lamp Tracking at Nighttime

Rear-lamp tracking at nighttime plays a momentous role in the advanced driver assistance system (ADAS), involving collision mitigation, automatic cruise control, automatic headlamp dimming, etc. Most of the existing tracking methods based on monocular camera leverage on color features. However, such tracking methods can be easily influenced by background clutter, illumination change, distance variation, and occlusion. In this paper, we propose an evolutionary adaptive rear-lamp tracking method at nighttime, in which a novel genetic algorithm powered by the probabilistic bitwise operation (PBO) is utilized. Also, to improve the robustness against various environments, a balanced fitness function is proposed by taking color information, symmetry, spatial relationship, and rigidity into account. Especially, a series of adaptive thresholds based on rear data in HSV color space is proposed to exploit color information reasonably with respect to our task. A strategy to deal with occlusion is also proposed, which relies on color information and rigidity. Moreover, to our knowledge, there is no publicly available dataset for rear-lamp tracking at nighttime. To fill the gap between the real-world application and the theoretical research, we create a novel dataset, which contains diverse traffic conditions at nighttime. The experimental results indicate that our method outperforms comparative online tracking methods in terms of success rate and center location error.

Binocular intelligent following robot based on YOLO-LITE

In order to solve the problem that the deep neural network model is large in scale, the calculation time is too long, and the real-time performance is severely limited when combined with embedded devices, so studied the intelligent follower robot system based on YOLO-LITE algorithm combined with Raspberry Pi 3B+. The system mainly includes camera processing, target detection and other modules. Obtained the internal and external parameters of the camera through calibration, and according to these parameters to correct the binocular camera. Recognized and located the target in each frame of image, calculated the distance from the camera to the target and the center location error, and driven the car to move. The experimental results show that the following car has excellent real-time performance, the average detection frame rate can reach 20Fps, and the average detection accuracy can reach more than 80%.

A quantitative attribute-based benchmark methodology for single-target visual tracking

In the past several years, various visual object tracking benchmarks have been proposed, and some of them have been used widely in numerous recently proposed trackers. However, most of the discussions focus on the overall performance, and cannot describe the strengths and weaknesses of the trackers in detail. Meanwhile, several benchmark measures that are often used in tests lack convincing interpretation. In this paper, 12 frame-wise visual attributes that reflect different aspects of the characteristics of image sequences are collated, and a normalized quantitative formulaic definition has been given to each of them for the first time. Based on these definitions, we propose two novel test methodologies, a correlation-based test and a weight-based test, which can provide a more intuitive and easier demonstration of the trackers’ performance for each aspect. Then these methods have been applied to the raw results from one of the most famous tracking challenges, the Video Object Tracking (VOT) Challenge 2017. From the tests, most trackers did not perform well when the size of the target changed rapidly or intensely, and even the advanced deep learning based trackers did not perfectly solve the problem. The scale of the targets was not considered in the calculation of the center location error; however, in a practical test, the center location error is still sensitive to the targets’ changes in size.

Multi-Feature Fusion in Particle Filter Framework for Visual Tracking

In this article, a particle filter based tracking algorithm is proposed to track a target in video with vivid and complex environments. The target is represented in feature space by both color distribution and KAZE features. Color distribution is selected for its robustness to target’s scale variation and partial occlusion. KAZE features are chosen for their ability to represent the target structure and also for their superior performance in feature matching. Fusion of these two features will lead to effective tracking as compared to other features due to their better representational abilities, under challenging conditions. The trajectory of the target is established using the particle filter algorithm based on similarity between the extracted features from the target and the probable candidates in the consecutive frames. For the color distribution model, Bhattacharya coefficient is used as a similarity metric whereas Nearest Neighbor Distance Ratio is used for matching of corresponding feature points in KAZE algorithm. The particle filter update model is based on kinematic motion equations and the weights on particles are governed by an equation fusing both the color and KAZE features. Centre Location Error, Average Tracking Accuracy and Tracking Success Rate are the performance metrics considered in the evaluation process. Also, the overlap success plot and precision plot is considered for performance evaluation. On the basis of these metrics and visual results obtained under different environment conditions: outdoor, occluding and underwater ones, the proposed tracking scheme performs significantly better than the contemporary feature-based iterative object tracking methods and even few of the learning-based algorithms.

Fast Scale Estimation Method in Object Tracking

Accurate and fast scale estimation of targets is a challenging research problem in visual object tracking. Most trackers employ an exhaustive scale search to estimate the target size. The exhaustive search strategy is computationally expensive and it struggles when encountered with large scale variations. In this paper, a scale adaptive method is proposed, which not only improves the tracking performance, but also greatly reduces the computational costs and improves the tracking speed. Based on the scale estimation method of SAMF, the original 7 fixed scale sizes were reduced to 3, and an adaptive scale size was added. Three fixed scales were used to determine the direction of scale change, and the APCE change rate of the current frame and the previous frame was used to control another adaptive scale size. Finally, the optimal scale estimation was determined. Additionally, we investigate the update strategy to further improve the tracking accuracy. Extensive experiments on OTB50, OTB100 and VOT-ST2019 datasets demonstrate that the proposed method can tackle challenging videos well compared with baseline tracker. On OTB, we obtain a gain of 7.0% in Distance Precision, and 18.8% in Centre Location Error on the selected 43 videos with scale variation attribute, and a mean gain of 6.2% in Precision and 4.6% in Success plots on OTB50, compared with the baseline tracker SAMF. Furthermore, the proposed approach improves the tracking speed by 34% in FPS compared with SAMF.

Automatic multiple moving humans detection and tracking in image sequences taken from a stationary thermal infrared camera

Several Particle Filter (PF)-based methods for human tracking in thermal IR image sequences have been proposed in the literature. Unfortunately, the majority of these methods are developed for tracking only a single human. Moreover, this human is manually pre-selected in the first frame of the image sequence, which is not practical for the real case of intelligent and efficient video surveillance system that needs tracking more than one human and without any external operator intervention. To contribute to addressing this need, in this paper, we propose a novel PF-based method that detects and tracks multiple moving humans using a thermal IR camera, without prior knowledge about their number and initial locations in the monitored scene. This method consists of three main parts. In the first one, all the moving objects are extracted from the image sequence by using the Gaussian Mixture Model (GMM) and then, for each extracted object, a combined shape, appearance, spatial and temporal-based similarity function that allows us to detect a human without any prior training of a mathematical model is calculated. The second part consists in tracking the human previously detected by using a PF and an adaptive combination of spatial, intensity, texture and motion velocity cues. In each cue, a model for the detected human is created, and when new observations arrive in the next frames, the similarity distances between each created model and the observed moving regions are calculated. The human tracking is achieved by combining individual similarity distances using adaptive weights, into a PF algorithm. The third part is devoted to detect and handle occlusions by using simple heuristic rules and grayscale Vertical Projection Histogram (VPH). Each part of the proposed method was separately tested on a set of real-world thermal IR image sequences containing background clutters, appearance and disappearance of multiple moving objects, occlusions, illumination and scales changes. A comparative study with several state-of-the-art methods has shown that the proposed method performs consistently better in terms of Center Location Error (CLE) and the Success Rate (SR), and it can also run at speed of about 15 Frame Per Second per human, which is considerably enough for real-time applications.

Accurate Tracking of Manoeuvring Target using Scale Estimation and Detection

Camera zoom operation and fast approaching/receding target causes scaling of acquired target in video frames. Fast moving target manifests in large inter-frame motion. In general, non-uniform background degrades performance of tracking algorithms. Fast Fourier transform (FFT)-based Correlation algorithms improve tracking in this scenario, but their applications is limited to small inter-frame motion. Increasing search region has implication on execution speed of the algorithms. Rapid target scaling, non-uniform background and large inter-frame motion of target hinder accurate and long term visual tracking. These challenges have been addressed for extended target tracking by augmenting fast discriminative scale space tracking (fDSST) algorithm with probable target location prediction and target detection. Localisation of fast motion has been achieved by applying fused outputs of Kalman filter and quadratic regression based prediction before applying fDSST. It has helped in accurate localisation of fast motion without increasing search region. In each frame, target location and size have been estimated using fDSST and further refined by target detection near this location. Smoothing and limiting of trajectory and size of detected target has enhanced tracking performance. Experimental results show considerable improvement of precision, success rate and centre location error tracking performance against state-of-the-art trackers in stringent conditions.

Salient Superpixel Visual Tracking with Graph Model and Iterative Segmentation

Visual object tracking is to locate an object of interest in a sequence of consecutive video frames, which is widely applied in many high-level computer vision tasks such as intelligent video surveillance and robotics. It is of great challenges for visual tracking methods to handle large target appearance variations caused by pose deformation, fast motion, occlusion, and surrounding environments in real-time videos. In this paper, inspired by human attention cognitive saliency model, we propose a visual tracking method based on salient superpixels which integrates the target appearance similarity and cognitive saliency, and helps to location inference and appearance model updating. The saliency of superpixel is detected by graph model and manifold ranking. We cluster the superpixels of the first four target boxes into a set corresponding to object foreground and model the target appearance with color descriptors. While tracking, the relevance is computed between the candidate superpixels and the target appearance set. We also propose an iterative threshold segmentation method to distinguish the foreground and background of superpixels based on saliency and relevance. To increase the accuracy of location inference, we explore particle filter in both confidence estimation and sampling procedures. We compared our method with the existing techniques in OTB100 dataset in terms of precision based on center location error and success rate based on overlap, and the experimental results show that our proposed method achieved substantially better performance. Promising results have shown that the proposed salient superpixel-based approach is effective to deformation, occlusion, and other challenges in object tracking.