Deep learning has been widely used in visual tracking due to strong feature extraction ability of convolutional neural network(CNN). Many trackers pre-train CNN primarily and fine-tune it during tracking, which could improve representation ability from off-line database and adjust to appearance variation of the interested object. However, since target information is limited, the network is likely to overfit to a single target state. In this paper, an update strategy composed of two modules is proposed. First, we fine-tune the pre-trained CNN using active learning that emphasizes the most discriminative data iteratively. Second, artificial convolutional features generated from empirical distribution are employed to train fully connected layers, which makes up the deficiency of training examples. Experiments evaluated on VOT2016 benchmark shows that our algorithm outperforms many state-of-the-art trackers.
Robust object tracking is a challenging task in computer vision due to interruptions such as deformation, fast motion and especially, occlusion of tracked object. When occlusions occur, image data will be unreliable and is insufficient for the tracker to depict the object of interest. Therefore, most trackers are prone to fail under occlusion. In this paper, an occlusion judgement and handling method based on segmentation of the target is proposed. If the target is occluded, the speed and direction of it must be different from the objects occluding it. Hence, the value of motion features are emphasized. Considering the efficiency and robustness of Kernelized Correlation Filter Tracking (KCF), it is adopted as a pre-tracker to obtain a predicted position of the target. By analyzing long-term motion cues of objects around this position, the tracked object is labelled. Hence, occlusion could be detected easily. Experimental results suggest that our tracker achieves a favorable performance and effectively handles occlusion and drifting problems.
Video tracking is a main field of computer vision, and TLD algorithm plays a key role in long-term tracking. However, the original TLD ignores the color features of patch in detection, and tracks the common points from grid, then, the tracking accuracy is limited to both of them. This paper presents a novel TLD algorithm with Harris corner and color moment to overcome this drawback. Instead of tracking common points, we screen more important points utilizing Harris corner to reject a half patches, these points are better able to show the object’s textural features. In addition, the color moment classifier replaces patch variance to reduce the errors of detection. The classifier compares mine-dimensional color moment vectors so that it can keep the TLD’s stable speed. Experiment has proved that our TLD tracks a more reliable position and higher ability without affecting the speed.
Visual tracking is important in computer vision. At present, although many algorithms of visual tracking have been proposed, there are still many problems which are needed to be solved, such as occlusion and frame speed. To solve these problems, this paper proposes a novel method which based on compressive tracking. Firstly, we make sure the occlusion happens if the testing result about image features by the classifiers is lower than a threshold value which is certain. Secondly, we mark the occluded image and record the occlusion region. In the next frame, we test both the classifier and the marked image. This algorithm makes sure the tracking is fast, and the result about solving occlusion is much better than other algorithms, especially compressive tracking.
Human action recognition and analysis is an active research topic in computer vision for many years. This paper presents a method to represent human actions based on trajectories consisting of 3D joint positions. This method first decompose action into a sequence of meaningful atomic actions (actionlets), and then label actionlets with English alphabets according to the Davies-Bouldin index value. Therefore, an action can be represented using a sequence of actionlet symbols, which will preserve the temporal order of occurrence of each of the actionlets. Finally, we employ sequence comparison to classify multiple actions through using string matching algorithms (Needleman-Wunsch). The effectiveness of the proposed method is evaluated on datasets captured by commodity depth cameras. Experiments of the proposed method on three challenging 3D action datasets show promising results.
KEYWORDS: Data compression, Information operations, Networks, Mobile communications, Data communications, Global system for mobile communications, Data acquisition, Data analysis, Lithium, Distance measurement
Collecting reliable and accurate MR data on time plays a vital role in the mobile communication network optimization. However, with the increment of the number of mobile users, network bandwidth cannot meet with mass transfer of MR. A high performance and high compression ratio GSM-MR compression algorithm is proposed to gain better transfer time. This algorithm utilizes two step sorting in order to reduce the distance between similar content, based on the analytic result about similarities of GSM-MR data sorting by different fields. Experimental results reveal that the algorithm does not only decrease compression consuming time, but also ascends compression ratio with the increment of the size of compression data.
Visual tracking is one of the significant research directions in computer vision. Although standard random ferns tracking method obtains a good performance for the random spatial arrangement of binary tests, the effect of the locality of image on ferns description ability are ignored and prevent them to describe the object more accurately and robustly. This paper proposes a novel spatial arrangement of binary tests to divide the bounding box into grids in order to keep more details of the image for visual tracking. Experimental results show that this method can improve tracking accuracy effectively.
Visual tracking is an important task in computer vision. Despite many researches have been done in this area, some problems remain. One of the problems is drifting. To handle the problem, a new appearance model update method based on a forward filtering backward smoothing particle filter is proposed in this paper. A smoothing of previous appearance model is performed by exploiting information of current frame instead of updating instantly in traditional tracking methods. It has been shown that smoothing based on future observations makes previous and current predictions more accurate, thus the appearance model update by our approach is more accurate. And at the same time, online tracking is achieved compared with some previous work in which the smoothing is done in an offline way. With the smoothing procedure, the tracker is more accurate and less likely to drift than traditional ones. Experimental results demonstrate the effectiveness of the proposed method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.