Limits...
Incremental structured dictionary learning for video sensor-based object tracking.

Xue M, Yang H, Zheng S, Zhou Y, Yu Z - Sensors (Basel) (2014)

Bottom Line: To formulate the training and classification process, a multiple linear classifier group based on a K-combined voting (KCV) function is proposed.As the dictionary evolves, the models are also trained to timely adapt the target appearance variation.We also illustrate its relay application in visual sensor networks.

View Article: PubMed Central - PubMed

Affiliation: Institute of Image Communication and Network Engineering, Shanghai Jiao Tong University, Shanghai 200240, China. xue@sjtu.edu.cn.

ABSTRACT
To tackle robust object tracking for video sensor-based applications, an online discriminative algorithm based on incremental discriminative structured dictionary learning (IDSDL-VT) is presented. In our framework, a discriminative dictionary combining both positive, negative and trivial patches is designed to sparsely represent the overlapped target patches. Then, a local update (LU) strategy is proposed for sparse coefficient learning. To formulate the training and classification process, a multiple linear classifier group based on a K-combined voting (KCV) function is proposed. As the dictionary evolves, the models are also trained to timely adapt the target appearance variation. Qualitative and quantitative evaluations on challenging image sequences compared with state-of-the-art algorithms demonstrate that the proposed tracking algorithm achieves a more favorable performance. We also illustrate its relay application in visual sensor networks.

No MeSH data available.


Single voting and random combined voting. The former scheme is a special case of the latter one with K = 1.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3958219&req=5

f4-sensors-14-03130: Single voting and random combined voting. The former scheme is a special case of the latter one with K = 1.

Mentions: We consider the discriminative observation modeling on a patch level. Given the patch-based coefficients, β, of candidate Xt are obtained and P candidates have been sampled, for the p-th candidate, p = 1, 2,…, P, a function called K-combined voting (KCV) is proposed to compute the score, S(p*), recording the times that p* is selected as the result by:(16)p(yt/Xt)=argmaxp*∑i=1cNS(pi*)(17)S(pi*)=1(18)pi*=maxp(1−α1+e−wt−1iβpC(i)+α1+e−w0iβpC(i))where , i = 1,2,…,CN refers to the i-th combinatory candidate given K > 1. When K = 1, it refers to the common single patch voting case, shown in Figure 4. refers to the support vector initially obtained for the i-th patch, while is vector generated at time t − 1. The K-combined voting can be seen as an efficient hierarchical generalization form of single voting. Based on each combination as the intermediate output, the final result provides a more comprehensive and neutral value within the sampled particles. In the case of drastic appearance variation, the random combined voting could provide more opportunities for the invariant patches to attend the voting calculation, so that it is more likely to obtain better results. Comparison between single voting and K-combined voting has been done as the proof in the next section. , are weight vectors of the i-th classifier learned at the first frame and time t − 1, and α is a constant.


Incremental structured dictionary learning for video sensor-based object tracking.

Xue M, Yang H, Zheng S, Zhou Y, Yu Z - Sensors (Basel) (2014)

Single voting and random combined voting. The former scheme is a special case of the latter one with K = 1.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3958219&req=5

f4-sensors-14-03130: Single voting and random combined voting. The former scheme is a special case of the latter one with K = 1.
Mentions: We consider the discriminative observation modeling on a patch level. Given the patch-based coefficients, β, of candidate Xt are obtained and P candidates have been sampled, for the p-th candidate, p = 1, 2,…, P, a function called K-combined voting (KCV) is proposed to compute the score, S(p*), recording the times that p* is selected as the result by:(16)p(yt/Xt)=argmaxp*∑i=1cNS(pi*)(17)S(pi*)=1(18)pi*=maxp(1−α1+e−wt−1iβpC(i)+α1+e−w0iβpC(i))where , i = 1,2,…,CN refers to the i-th combinatory candidate given K > 1. When K = 1, it refers to the common single patch voting case, shown in Figure 4. refers to the support vector initially obtained for the i-th patch, while is vector generated at time t − 1. The K-combined voting can be seen as an efficient hierarchical generalization form of single voting. Based on each combination as the intermediate output, the final result provides a more comprehensive and neutral value within the sampled particles. In the case of drastic appearance variation, the random combined voting could provide more opportunities for the invariant patches to attend the voting calculation, so that it is more likely to obtain better results. Comparison between single voting and K-combined voting has been done as the proof in the next section. , are weight vectors of the i-th classifier learned at the first frame and time t − 1, and α is a constant.

Bottom Line: To formulate the training and classification process, a multiple linear classifier group based on a K-combined voting (KCV) function is proposed.As the dictionary evolves, the models are also trained to timely adapt the target appearance variation.We also illustrate its relay application in visual sensor networks.

View Article: PubMed Central - PubMed

Affiliation: Institute of Image Communication and Network Engineering, Shanghai Jiao Tong University, Shanghai 200240, China. xue@sjtu.edu.cn.

ABSTRACT
To tackle robust object tracking for video sensor-based applications, an online discriminative algorithm based on incremental discriminative structured dictionary learning (IDSDL-VT) is presented. In our framework, a discriminative dictionary combining both positive, negative and trivial patches is designed to sparsely represent the overlapped target patches. Then, a local update (LU) strategy is proposed for sparse coefficient learning. To formulate the training and classification process, a multiple linear classifier group based on a K-combined voting (KCV) function is proposed. As the dictionary evolves, the models are also trained to timely adapt the target appearance variation. Qualitative and quantitative evaluations on challenging image sequences compared with state-of-the-art algorithms demonstrate that the proposed tracking algorithm achieves a more favorable performance. We also illustrate its relay application in visual sensor networks.

No MeSH data available.