Automatic continuous tracking of objects involved in a construction project is required for such tasks as productivity assessment, unsafe behavior recognition, and progress monitoring. Many computer-vision-based tracking approaches have been investigated and successfully tested on construction sites; however, their practical applications are hindered by the tracking accuracy limited by the dynamic, complex nature of construction sites (i.e. clutter with background, occlusion, varying scale and pose). To achieve better tracking performance, a novel deep-learning-based tracking approach called the Multi-Domain Convolutional Neural Networks (MD-CNN) is proposed and investigated. The proposed approach consists of two key stages: 1) multi-domain representation of learning; and 2) online visual tracking. To evaluate the effectiveness and feasibility of this approach, it is applied to a metro project in Wuhan China, and the results demonstrate good tracking performance in construction scenarios with complex background. The average distance error and F-measure for the MDNet are 7.64 pixels and 67, respectively. The results demonstrate that the proposed approach can be used by site managers to monitor and track workers for hazard prevention in construction sites.

Visual object tracking is always a challenging task in computer vision. During the tracking, the shape and appearance of the target may change greatly, and because of the lack of sufficient training samples, most of the online learning tracking algorithms will have performance bottlenecks. In this paper, an improved real-time algorithm based on deep learning features is proposed, which combines multi-feature fusion, multi-scale estimation, adaptive updating of target model and re-detection after target loss. The effectiveness and advantages of the proposed algorithm are proved by a large number of comparative experiments with other excellent algorithms on large benchmark datasets.

In daily reality, people often pay attention to several objects that change positions while being observed. In the laboratory, this process is investigated by a phenomenon known as multiple object tracking (MOT) which is a task that evaluates attentive tracking performance. Recent findings suggest that the attentional set for multiple moving objects whose depth changes in three dimensions from one plane to another is influenced by the initial configuration of the objects. When tracking objects, it is difficult for people to expand their attentional set to multiple-depth planes once attention has been focused on a single plane. However, less is known about people contracting their attentional set from multiple-depth planes to a single-depth plane. In two experiments, we examined tracking accuracy when four targets or four distractors, which were initially distributed on two planes, come together on one of the planes during an MOT task. The results from this study suggest that people have difficulty changing the depth range of their attention during attentive tracking, and attentive tracking performance depends on the initial attentional set based on the configuration prior to attentive tracking.

Deep neural networks have achieved great success in visual tracking by learning a generic representation and leveraging large amounts of training data to improve performance. Most generic object trackers are trained from scratch online and do not benefit from a large number of videos available for offline training. We present a real-time generic object tracker capable of incorporating temporal information into its model, learning from many examples offline and quickly updating online. During the training process, the pre-trained weight of convolution layer is updated lagging behind, and the input video sequence length is gradually increased for fast convergence. Furthermore, only the hidden states in recurrent network are updated to guarantee the real-time tracking speed. The experimental results show that the proposed tracking method is capable of tracking objects at 150 fps with higher predicting overlap rate, and achieves more robustness in multiple benchmarks than state-of-the-art performance.

Pre-trained CNNs on ImageNet have been widely used in object tracking for feature extraction. However, due to the domain mismatch between image classification and object tracking, the submergence of the target-specific features by noise largely decreases the expression ability of the convolutional features, resulting in an inefficient tracking. In this paper, we propose a robust tracking algorithm with low-dimensional target-specific feature extraction. First, a novel cascaded PCA module is proposed to have an explicit extraction of the low-dimensional target-specific features, which makes the new appearance model more effective and efficient. Next, a fast particle filter process is raised to further accelerate the whole tracking pipeline by sharing convolutional computation with a ROI-Align layer. Moreover, a classification-score guided scheme is used to update the appearance model for adapting to target variations while at the same time avoiding the model drift that caused by the object occlusion. Experimental results on OTB100 and Temple Color128 show that, the proposed algorithm has achieved a superior performance among real-time trackers. Besides, our algorithm is competitive with the state-of-the-art trackers in precision while runs at a real-time speed.

Instance features based deep learning methods prompt the performances of high speed object tracking systems by directly comparing target with its template during training and tracking. However, from the perspective of human vision system, prior knowledge of target also plays key role during the process of tracking. To integrate both semantic knowledge and instance features, we propose a convolutional network based object tracking framework to simultaneously output bounding boxes based on different prior knowledge as well as confidences of corresponding Assumptions. Experimental results show that our proposed approach retains both higher accuracy and efficiency than other leading methods on tracking tasks covering most daily objects.

On-line object tracking (OLOT) has been a core technology in computer vision, and its importance has been increasing rapidly. Because this technology is utilized for battery-operated products, energy consumption must be minimized. This paper describes a method of adaptive frame-rate optimization to satisfy that requirement. An energy trade-off occurs between image capturing and object tracking. Therefore, the method optimizes the frame-rate based on always changed object speed for minimizing the total energy while taking into account the trade-off. Simulation results show a maximum energy reduction of 50.0%, and an average reduction of 35.9% without serious tracking accuracy degradation.

Most state-of-the-art discriminative tracking approaches are based on either template appearance models or statistical appearance models. Despite template appearance models have shown excellent performance, they perform poorly when the target appearance changes rapidly. In contrast, statistic appearance models are insensitive to fast target state changes, but they yield inferior tracking results in challenging scenarios such as illumination variations and background clutters. In this paper, we propose an adaptive object tracking approach with complementary models based on template and statistical appearance models. Both of these models are unified via our novel combination strategy. In addition, we introduce an efficient update scheme to improve the performance of our approach. Experimental results demonstrate that our approach achieves superior performance at speeds that far exceed the frame-rate requirement on recent tracking benchmarks.

In general, effective integrating the advantages of different trackers can achieve unified performance promotion. In this work, we study the integration of multiple correlation filter (CF) trackers; propose a novel but simple tracking integration method that combines different trackers in filter level. Due to the variety of their correlation filter and features, there is no comparability between different CF tracking results for tracking integration. To tackle this, we propose twofold CF to unify these various response maps so that the results of different tracking algorithms can be compared, so as to boost the tracking performance like ensemble learning. Experiment of two CF methods integration on the data sets OTB demonstrates that the proposed method is effective and promising.

Tracking-by-detection methods consider tracking task as a continuous detection problem applied over video frames. Modern tracking-by-detection trackers have online learning ability; the update stage is essential because it determines how to modify the classifier inherent in a tracker. However, most trackers search for the target within a fixed region centered at the previous object position; thus, they lack spatiotemporal consistency. This becomes a problem when the tracker detects an incorrect object during short-term occlusion. In addition, the scale of the bounding box that contains the target object is usually assumed not to change. This assumption is unrealistic for long-term tracking, where the scale of the target varies as the distance between the target and the camera changes. The accumulation of errors resulting from these shortcomings results in the drift problem, i.e. drifting away from the target object. To resolve this problem, we present a drift-free, online learning-based tracking-by-detection method using a single static camera. We improve the latent structured support vector machine (SVM) tracker by designing a more robust tracker update step by incorporating two Kalman filter modules: the first is used to predict an adaptive search region in consideration of the object motion; the second is used to adjust the scale of the bounding box by accounting for the background model. We propose a hierarchical search strategy that combines Bhattacharyya coefficient similarity analysis and Kalman predictors. This strategy facilitates overcoming occlusion and increases tracking efficiency. We evaluate this work using publicly available videos thoroughly. Experimental results show that the proposed method outperforms the state-of-the-art trackers.

In this letter, we propose a robust and fast tracking framework by combining local and global appearance models to cope with partial occlusion and pose variations. The global appearance model is represented by a correlation filter to efficiently estimate the movement of the target and the local appearance model is represented by local feature points to handle partial occlusion and scale variations. Then global and local appearance models are unified via the Bayesian inference in our tracking framework. We experimentally demonstrate the effectiveness of the proposed method in both terms of accuracy and time complexity, which takes 12ms per frame on average for benchmark datasets.

In this study, we address the problem of online RGB-D tracking which confronted with various challenges caused by deformation, occlusion, background clutter, and abrupt motion. Various trackers have different strengths and weaknesses, and thus a single tracker can merely perform well in specific scenarios. We propose a 3D tracker-level fusion algorithm (TLF3D) which enhances the strengths of different trackers and suppresses their weaknesses to achieve robust tracking performance in various scenarios. The fusion result is generated from outputs of base trackers by optimizing an energy function considering both the 3D cube attraction and 3D trajectory smoothness. In addition, three complementary base RGB-D trackers with intrinsically different tracking components are proposed for the fusion algorithm. We perform extensive experiments on a large-scale RGB-D benchmark dataset. The evaluation results demonstrate the effectiveness of the proposed fusion algorithm and the superior performance of the proposed TLF3D tracker against state-of-the-art RGB-D trackers.

This paper proposes an on-line rigid object tracking framework via discriminative object appearance modeling and learning. Strong classifiers are combined with 2D scale-rotation invariant local features to treat tracking as a keypoint matching problem. For on-line boosting, we correspond a Gaussian mixture model (GMM) to each weak classifier and propose a GMM-based classifying mechanism. Meanwhile, self-organizing theory is applied to perform automatic clustering for sequential updating. Benefiting from the invariance of the SURF feature and the proposed on-line classifying technique, we can easily find reliable matching pairs and thus perform accurate and stable tracking. Experiments show that the proposed method achieves better performance than previously reported trackers.

In this paper, a multiple-object tracking approach in large-scale scene is proposed based on visual sensor network. Firstly, the object detection is carried out by extracting the HOG features. Then, object tracking is performed based on an improved particle filter method. On the one hand, a kind of temporal and spatial dynamic model is designed to improve the tracking precision. On the other hand, the cumulative error generated from evaluating particles is eliminated through an appearance model. In addition, losses of the tracking will be incurred for several reasons, such as occlusion, scene switching and leaving. When the object is in the scene under monitoring by visual sensor network again, object tracking will continue through object re-identification. Finally, continuous multiple-object tracking in large-scale scene is implemented. A database is established by collecting data through the visual sensor network. Then the performances of object tracking and object re-identification are tested. The effectiveness of the proposed multiple-object tracking approach is verified.

Object tracking is one of the fundamental problems in computer vision. However, there is still a need to improve the overall capability in various tracking circumstances. In this letter, a patches-collaborative compressive tracking (PCCT) algorithm is presented. Experiments on various challenging benchmark sequences demonstrate that the proposed algorithm performs favorably against several state-of-the-art algorithms.

A non-linear extension of generalized hyperplane approximation (GHA) method is introduced in this letter. Although GHA achieved a high-confidence result in motion parameter estimation by utilizing the supervised learning scheme in histogram of oriented gradient (HOG) feature space, it still has unstable convergence range because it approximates the non-linear function of regression from the feature space to the motion parameter space as a linear plane. To extend GHA into a non-linear regression for larger convergence range, we derive theoretical equations and verify this extension's effectiveness and efficiency over GHA by experimental results.

This paper proposes a novel object tracking method via online boosting. The on-line boosting technique is combined with local features to treat tracking as a keypoint matching problem. First, We improve matching reliability by exploiting the statistical repeatability of local features. In addition, we propose 2D scale-rotation invariant quasi-keypoint matching to further improve matching efficiency. Benefiting from SURF feature's statistical repeatability and the complementary quasi-keypoint matching technique, we can easily find reliable matching pairs and thus perform accurate and stable tracking. Experimental results show that the proposed method achieves better performance compared with previously reported trackers.

The problem of identifying moving objects in a video recording produced by a range sensor camera is due to the limited information available for classifying different objects. On the other hand, the infrared signal from a range sensor camera is more robust for extreme luminance intensity when the monitored area has light conditions that are too bright or too dark. This paper proposes a method of detection and tracking moving objects in image sequences captured by stationary range sensor cameras. Here, the depth information is utilized to correctly identify each of detected objects. Firstly, camera calibration and background subtraction are performed to separate the background from the moving objects. Next, a 2D projection mapping is performed to obtain the location and contour of the objects in the 2D plane. Based on this information, graph matching is performed based on features extracted from the 2D data, namely object position, size and the behavior of the objects. By observing the changes in the number of objects and the objects' position relative to each other, similarity matching is performed to track the objects in the temporal domain. Experimental results show that by using similarity matching, object identification can be correctly achieved even during occlusion.

This paper presents an automatic method to track soccer players in soccer video recorded from a single camera where the occurrence of pan-tilt-zoom can take place. The automatic object tracking is intended to support texture extraction in a free viewpoint video authoring application for soccer video. To ensure that the identity of the tracked object can be correctly obtained, background segmentation is performed and automatically removes commercial billboards whenever it overlaps with the soccer player. Next, object tracking is performed by an attribute matching algorithm for all objects in the temporal domain to find and maintain the correlation of the detected objects. The attribute matching process finds the best match between two objects in different frames according to their pre-determined attributes: position, size, dominant color and motion information. Utilizing these attributes, the experimental results show that the tracking process can handle occlusion problems such as occlusion involving more than three objects and occluded objects with similar color and moving direction, as well as correctly identify objects in the presence of camera movements.

This paper seeks to improve power-performance efficiency of embedded systems by the use of dynamic reconfiguration. Programmable logic devices (PLDs) have the competence to optimize the power consumption by the use of partial and/or dynamic reconfiguration. It is a non-exclusive approach, which can use other power-reduction techniques simultaneous, and thus it is applicable to a myriad of systems. The power-performance improvement by dynamic reconfiguration was evaluated through an augmented reality system that translates Japanese into English. It is a wearable and mobile system with a head-mounted display (HMD). In the system, the computing core detects a Japanese word from an input video frame and the translated term will be output to the HMD. It includes various image processing approaches such as pattern recognition and object tracking, and these functions run sequentially. The system does not need to prepare all functions simultaneously, which provides a function by reconfiguration only when it is needed. In other words, by dynamic reconfiguration, the spatiotemporal module-based pipeline can introduce the reduction of its circuit amount and power consumption compared to the naive approach. The approach achieved marked improvements; the computational speed was the same but the power consumption was reduced to around $rac{1}{6}$.