面向多目標視頻的出生強度估計方法
Date Issued
2014
ISSN
1003-9775
Citation
計算機輔助設計與圖形學學報, 2014年, 第26卷, 第12期, 第2223-2231頁.
Type
Peer Reviewed Journal Article
Abstract
針對多目標視訊追蹤中的新生目標出生強度估計問題,提出一種有效的基於熵分佈和覆蓋率的方法. 此方法利用前一時刻所獲目標狀態及測量值對出生強度進行初始化,再利用當前時刻所獲測量值對出生強度進行更新. 在更新階段,首先選取僅依賴權值的負指數分佈作為出生強度的時間無關的雜訊分量;然後透過計算出生強度與對應測量值間的覆蓋率對出生強度權值進行再次更新,進一步濾除權值 小於給定閾值的雜訊分量。實驗結果顯示,文中方法有效地降低了雜訊成分的影響,提高了多目標視訊追蹤的準確率.
In this paper,an effective birth intensity estimation method that based on entropy distribution and coverage rate is proposed for multi-target video tracking.The birth intensity is first initialized according to the previously obtained target states and measurements.The currently obtained measurements are then used to update the initialized birth intensity.In the updating stage,the negative exponent entropy distribution that depends on the weight is first selected as the prior distribution of the birth intensity.By doing so,the components within the birth intensity those are irrelevant to the measurements could be regarded as noises and should be removed.The coverage rate between each birth intensity component and the corresponding measurement is then computed to further eliminate those components whose weights are smaller than the given threshold.Experiments on noisy video sequences are conducted to show that the proposed birth intensity estimation method can effectively eliminate the noises and finally improve the tracking accuracy.
In this paper,an effective birth intensity estimation method that based on entropy distribution and coverage rate is proposed for multi-target video tracking.The birth intensity is first initialized according to the previously obtained target states and measurements.The currently obtained measurements are then used to update the initialized birth intensity.In the updating stage,the negative exponent entropy distribution that depends on the weight is first selected as the prior distribution of the birth intensity.By doing so,the components within the birth intensity those are irrelevant to the measurements could be regarded as noises and should be removed.The coverage rate between each birth intensity component and the corresponding measurement is then computed to further eliminate those components whose weights are smaller than the given threshold.Experiments on noisy video sequences are conducted to show that the proposed birth intensity estimation method can effectively eliminate the noises and finally improve the tracking accuracy.
Loading...
Availability at HKSYU Library
