A Biologically Inspired Appearance Model for Robust Visual Tracking

Zhang, Shengping; Lan, Xiangyuan; Yao, Hongxun; Zhou, Huiyu; Tao, Dacheng; Li, Xuelong

doi:10.1109/tnnls.2016.2586194

Cited by 86 publications

(42 citation statements)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The initial appearance is preserved in a network model if the target samples drawn from the first frame are put into consideration when updated, which is helpful for target re-detection after its reappearance after temporary disappearance. Pessimistically believed in [46], from the author's point of view, only the positive sample from the first frame is completely reliable, whereas contamination and decision mistake must exist in other frames to some extent, which is also deemed to be true in [47] that there must exist error a bit or too much in each frame, except in the first one. However, optimistically speaking, thanks to the close appearances from the two adjacent frames, a trend of the variation can be foreseen within a small period (no above than three frames); therefore, there exists a high confidence of making sure whether the tracking result is responsible.…”

Section: Update Strategy Of Neural Network Modelsmentioning

confidence: 99%

Model Update Strategies about Object Tracking: A State of the Art Review

et al. 2019

View full text Add to dashboard Cite

Object tracking has always been an interesting and essential research topic in the domain of computer vision, of which the model update mechanism is an essential work, therefore the robustness of it has become a crucial factor influencing the quality of tracking of a sequence. This review analyses on recent tracking model update strategies, where target model update occasion is first discussed, then we give a detailed discussion on update strategies of the target model based on the mainstream tracking frameworks, and the background update frameworks are discussed afterwards. The experimental performances of the trackers in recent researches acting on specific sequences are listed in this review, where the superiority and some failure cases on each of them are discussed, and conclusions based on those performances are then drawn. It is a crucial point that design of a proper background model as well as its update strategy ought to be put into consideration. A cascade update of the template corresponding to each deep network layer based on the contributions of them to the target recognition can also help with more accurate target location, where target saliency information can be utilized as a tool for state estimation.Diverse variations regarding the target usually occur in the process of tracking, i.e., variations arise from changes of the outside environment, such as view angle, camera orientation, environmental illumination, etc., and inherent changes of the object, such as self-rotation, self-deformation, and self-variation of target appearance; therefore, a tracker with more robust capacity has to be designed, whose framework structure and sample learning strategy are of key importance, which guarantees its real-time and accuracy. Consequently, researching an update strategy with higher robustness and efficiency has been of greater essence.Object tracking framework can be usually typed into two categories: generative frameworks and discriminative ones, where, for the former framework, i.e., particle filter, sparse coding, linear predictions [5,6], Kalman filter, etc., target and background models are established at the beginning and the features of them are extracted for the search of similar target or background features in succeeding frame images to iteratively locate the target; The latter, i.e., deep neural networks, correlation filter, random forest, feature bagging [7], etc., gets the object location by drawing candidate target patches within a region and then select one that is distinguished from given background patches. With the progress of researches on machine learning and deep learning tracking frameworks, the model update has become a widely concerned part in recent researches. A good update mechanism is a crucial respect measuring the reliability of a tracker. On the one hand, template models of the target and background should be constantly updated to catch up with the their variation, which is a fundamental requirement of model adaptation. On the other hand, the parameter model must be adjuste...

show abstract

Section: Update Strategy Of Neural Network Modelsmentioning

confidence: 99%

Model Update Strategies about Object Tracking: A State of the Art Review

et al. 2019

View full text Add to dashboard Cite

show abstract

“…The former focuses on modeling the appearance of the tracked target and then finds the candidate that is the most similar to the target template as the tracking result. The representative methods include those trackers based on sparse representation [23][24][25][26][27][28][29]. In [29], sparse coding is used to extract features from sampled patches.…”

Section: Related Workmentioning

confidence: 99%

“…Since features from different cues describe the tracked target from different aspects, more robust tracking results can be obtained when multi-cue features are used. In [23], a biologically inspired appearance model is proposed to model target appearance, which is also based on features extracted using sparse coding.…”

Section: Related Workmentioning

confidence: 99%

Visual Tracking Based on Discriminative Compressed Features

Liu

Wang

2018

Advances in Multimedia

View full text Add to dashboard Cite

Visual tracking is a challenging research topic in the field of computer vision with many potential applications. A large number of tracking methods have been proposed and achieved designed tracking performance. However, the current state-of-the-art tracking methods still can not meet the requirements of real-world applications. One of the main challenges is to design a good appearance model to describe the target's appearance. In this paper, we propose a novel visual tracking method, which uses compressed features to model target's appearances and then uses SVM to distinguish the target from its background. The compressed features were obtained by the zero-tree coding on multiscale wavelet coefficients extracted from an image, which have both the low dimensionality and discriminate ability and therefore ensure to achieve better tracking results. The experimental comparisons with several stateof-the-art methods demonstrate the superiority of the proposed method.

show abstract

“…Later, that model is used to estimate affinities between the objects across different frames. The approaches following this pipeline exploit several distinct types of representation models, including; appearance models [10], [11], [12], [13], motion models [14], [15], [16], [17], and composite models [9], [18], [19]. The appearance models focus on computing easy-to-track object features that encode appearances of local regions of objects or their bounding boxes [20], [21], [22].…”

Section: Introductionmentioning

confidence: 99%

Deep Affinity Network for Multiple Object Tracking

Sun

Akhtar

Song

et al. 2019

IEEE Trans. Pattern Anal. Mach. Intell.

250

140

View full text Add to dashboard Cite

Multiple Object Tracking (MOT) plays an important role in solving many fundamental problems in video analysis and computer vision. Most MOT methods employ two steps: Object Detection and Data Association. The first step detects objects of interest in every frame of a video, and the second establishes correspondence between the detected objects in different frames to obtain their tracks. Object detection has made tremendous progress in the last few years due to deep learning. However, data association for tracking still relies on hand crafted constraints such as appearance, motion, spatial proximity, grouping etc. to compute affinities between the objects in different frames. In this paper, we harness the power of deep learning for data association in tracking by jointly modeling object appearances and their affinities between different frames in an end-to-end fashion. The proposed Deep Affinity Network (DAN) learns compact, yet comprehensive features of pre-detected objects at several levels of abstraction, and performs exhaustive pairing permutations of those features in any two frames to infer object affinities. DAN also accounts for multiple objects appearing and disappearing between video frames. We exploit the resulting efficient affinity computations to associate objects in the current frame deep into the previous frames for reliable on-line tracking. Our technique is evaluated on popular multiple object tracking challenges MOT15, MOT17 and UA-DETRAC. Comprehensive benchmarking under twelve evaluation metrics demonstrates that our approach is among the best performing techniques on the leader board for these challenges. The open source implementation of our work is available at https://github.com/shijieS/SST.git.

show abstract

A Biologically Inspired Appearance Model for Robust Visual Tracking

Cited by 86 publications

References 66 publications

Model Update Strategies about Object Tracking: A State of the Art Review

Model Update Strategies about Object Tracking: A State of the Art Review

Visual Tracking Based on Discriminative Compressed Features

Deep Affinity Network for Multiple Object Tracking

Contact Info

Product

Resources

About