The template update problem

Matthews, LM; Ishikawa, Takahiro; Baker, Simon

doi:10.1109/tpami.2004.16

Cited by 675 publications

(118 citation statements)

References 12 publications

Supporting

Mentioning

103

Contrasting

Unclassified

Order By: Relevance

“…Looking at the models in table 3 used by the track validation algorithm, it is apparent that some of them do not accurately represent the target: the tracker was badly placed at the moment of reinitialisation, but the forwards and backwards trajectories matched sufficiently well for track validation to occur. Scenario 5 demonstrates this effect, sometimes called the "template update problem": each successive model drifts further away from its target [12]. It can be seen that once such an inaccuracy has arisen, it is likely to remain present in the subsequent models.…”

Section: Resultsmentioning

confidence: 99%

“…Firstly, it allows us to follow objects whose appearance changes drastically over time: table 3 contains many examples of lighting changes, articulation and out-of-plane rotation, which make tracking a person through an entire sequence using only a single model very challenging. The approach can therefore be regarded as a form of "unsupervised model building" [12], where the model adapts to changes in the target's appearance without manual intervention. Secondly, even when our algorithm is unable to track an object for the full length of a video clip, it is providing valuable information to the higher-level processes that invoked it.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Track validation using gradient-based normalised cross-correlation

Caulfield

Dawson‐Howe

2011

Procedings of the British Machine Vision Conference 2011

View full text Add to dashboard Cite

We develop a gradient-based normalised cross-correlation tracker that is as robust as brute-force template matching while being significantly more computationally efficient. The technique serves as the basis of our track validation algorithm: by tracking an object forwards in time, reinitialising at the end of the sequence and then tracking backwards in time, we can determine whether or not the object has been followed correctly -the forwards and backwards trajectories will be very different for tracking failures. If such a failure occurs, we iteratively attempt to validate shorter portions of the video sequence until validation is achieved. The algorithm provides a means of determining whether or not an object was tracked successfully without the need for ground truth data.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Track validation using gradient-based normalised cross-correlation

Caulfield

Dawson‐Howe

2011

Procedings of the British Machine Vision Conference 2011

View full text Add to dashboard Cite

show abstract

“…25). For incremental tracking at the frame rates of the current study, the primary source of error will often be for displacements which are much less than the pixel length.…”

Section: Discussionmentioning

confidence: 99%

Temporal regularization of ultrasound-based liver motion estimation for image-guided radiation therapy

2015

View full text Add to dashboard Cite

Purpose:Ultrasound‐based motion estimation is an expanding subfield of image‐guided radiation therapy. Although ultrasound can detect tissue motion that is a fraction of a millimeter, its accuracy is variable. For controlling linear accelerator tracking and gating, ultrasound motion estimates must remain highly accurate throughout the imaging sequence. This study presents a temporal regularization method for correlation‐based template matching which aims to improve the accuracy of motion estimates.Methods:Liver ultrasound sequences (15–23 Hz imaging rate, 2.5–5.5 min length) from ten healthy volunteers under free breathing were used. Anatomical features (blood vessels) in each sequence were manually annotated for comparison with normalized cross‐correlation based template matching. Five sequences from a Siemens Acuson™ scanner were used for algorithm development (training set). Results from incremental tracking (IT) were compared with a temporal regularization method, which included a highly specific similarity metric and state observer, known as the α–β filter/similarity threshold (ABST). A further five sequences from an Elekta Clarity™ system were used for validation, without alteration of the tracking algorithm (validation set).Results:Overall, the ABST method produced marked improvements in vessel tracking accuracy. For the training set, the mean and 95th percentile (95%) errors (defined as the difference from manual annotations) were 1.6 and 1.4 mm, respectively (compared to 6.2 and 9.1 mm, respectively, for IT). For each sequence, the use of the state observer leads to improvement in the 95% error. For the validation set, the mean and 95% errors for the ABST method were 0.8 and 1.5 mm, respectively.Conclusions:Ultrasound‐based motion estimation has potential to monitor liver translation over long time periods with high accuracy. Nonrigid motion (strain) and the quality of the ultrasound data are likely to have an impact on tracking performance. A future study will investigate spatial uniformity of motion and its effect on the motion estimation errors.

show abstract

“…Tracked object location, size and speed are some informative features used in event detection. Region tracking approaches (Matthews et al, 2004) might be less useful recorded for analysing behaviour patterns. Considering the above, here the object tracking methodology utilises the tracking principle proposed by Yang et al (2005) together with the improvements in Herath et al (2014), which is capable of detecting object-events such as merges and splits.…”

Section: Journal Of the National Science Foundation Of Sri Lanka 44(4)mentioning

confidence: 99%

Online tracking and event clustering for vision systems

Perera¹,

Herath²,

Fernando³

et al. 2016

J. Natn. Sci. Foundation Sri Lanka

View full text Add to dashboard Cite

This paper proposes a comprehensive method for online-event clustering in videos. Adaptive Gaussian mixture of foreground and they were modelled using the time series modelling technique. A cross-substitution based model comparison method was employed to compare the disparity between events. Spectral clustering (SC) was utilised to cluster events, and methods for SC initial parameter selection have been proposed. A method for cluster identity assignment in consecutive clustering iterations is also utilised to handle the evolving nature of the unsupervised learning methodology adopted. The proposed method is capable of producing reliable clustering results online, amidst a number of complications including dynamic backgrounds, object shadows, camera distortions, sudden foreground bursts and inter-object interactions.Event clustering, foreground estimation, trajectory modelling. INTRODUCTIONHumans are capable of identifying similarities and dissimilarities between the events they perceive. The objective of event clustering is to produce a set of event clusters to follow human heuristics through an automated process. Such systems are widely applied in (Gowsikhaa et al., 2014), bio vision applications (Jiang et al., 2013; Aqel et al., 2016) and abnormal event detection (Archetti et al., 2006 ; Guo et al., 2013;Ranjith et al., 2015). of activities performed by an object (Porikli & Haga, 2004;Utsumi et al., 2013). An event is described using a set of temporal features of the relevant object such as its location, velocity and size. Therefore, obtaining various feature variations of objects under consideration becomes one of the primary exercises in event clustering. In order to construct the necessary feature space, the desired foreground objects need to be segmented and tracked throughout the video. Therefore, any event detection practice comprises a series of sub processes, namely, foreground estimation, construction and clustering. The paper proposes a comprehensive methodology, which addresses each of these sub processes in a sequential manner to arrive at a robust event clustering system. Most of the previously proposed foreground estimation methods utilise the principle of background subtraction (Zang & Klette, 2004;Qin et al., 2013). Employment of a single background has long gone obsolete due to its inability to capture gradual changes in the background. Rahman et al. (2010) and Yang et al. (2005) have proposed remedies to absorb the background illumination change into the background reference model. But both these methods are ineffective when external objects are introduced to the background. The method proposed in Stauffer and Grimson (2000) is capable of tackling the issue of gradual background changes by modelling each pixel as a mixture of Gaussian distributions.

show abstract

The template update problem

Cited by 675 publications

References 12 publications

Track validation using gradient-based normalised cross-correlation

Track validation using gradient-based normalised cross-correlation

Temporal regularization of ultrasound-based liver motion estimation for image-guided radiation therapy

Online tracking and event clustering for vision systems

Contact Info

Product

Resources

About