【摘要】：針對單目靜態(tài)圖像中姿態(tài)估計方法易受遮擋干擾的問題,提出基于部位遮擋級別的可形變姿態(tài)估計方法.首先定義遮擋級別為人體部位的被遮擋程度,其通過計算部位遮擋比例和部位方向獲得;然后根據(jù)遮擋級別為每個部位建立對應(yīng)級別的部位檢測器,并給出基于遮擋級別的部位間形變模型;最后依據(jù)以上2個模型的總體匹配得分,獲得最合理的人體姿態(tài).在標準數(shù)據(jù)集IP和LSP上的實驗結(jié)果表明,該方法提高了姿態(tài)估計的整體準確率,特別是減少了有遮擋情況下的部位誤匹配問題.
[Abstract]:A deformable attitude estimation method based on occlusion level is proposed to solve the problem that the attitude estimation method in monocular still images is vulnerable to occlusion interference. Firstly, the degree of occlusion is defined as the degree of occlusion of human body, which is obtained by calculating the proportion and direction of occlusion. Then according to the occlusion level, the corresponding position detector is established for each position, and the deformation model between the parts based on the occlusion level is given. Finally, according to the overall matching score of the above two models, the most reasonable posture of the human body is obtained. The experimental results on the standard data sets IP and LSP show that the proposed method improves the overall accuracy of attitude estimation, especially reduces the mismatch problem in the case of occlusion.
【作者單位】： 東北大學(xué)計算機科學(xué)與工程學(xué)院;沈陽工程學(xué)院信息學(xué)院;沈陽航空航天大學(xué)計算機學(xué)院;
【基金】：國家自然科學(xué)基金(61170185) 遼寧省博士啟動基金(20121034) 遼寧省教育廳科學(xué)研究一般項目(L2014070,L2015368,L201602)
【分類號】：TP391.41

【相似文獻】

相關(guān)期刊論文 前10條

1 韓貴金;朱虹;;一種基于圖結(jié)構(gòu)模型的人體姿態(tài)估計算法[J];計算機工程與應(yīng)用;2013年14期

2 王浩;劉則芬;方寶富;陳金金;;基于約束樹形圖結(jié)構(gòu)外觀模型的人體姿態(tài)估計[J];計算機科學(xué);2014年03期

3 蘇延超;艾海舟;勞世z，

本文編號：2340945