计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2015年
16期
17-22
,共6页
Hausdorff距离%支持向量机%手势识别%手势分割%序贯最小优化
Hausdorff距離%支持嚮量機%手勢識彆%手勢分割%序貫最小優化
Hausdorff거리%지지향량궤%수세식별%수세분할%서관최소우화
Hausdorff distance%Support Vector Machine%gesture segmentation%gesture recognition%Sequential Minimal Optimization
手势识别技术作为最有前景的一种自然人机交互模式已经成功应用于一些领域。可靠的手势识别技术多依赖特定的硬件实现，而这种自然交互模式的普及需要自然环境下基于普通摄像机的通用手势识别技术。研究了在普通摄像机下对各种复杂背景、不同光照条件的静态手势的分割和识别技术。首先采用一种邻域变换算法，克服不同光照强度对分割的影响，然后提出一种求最小平均Hausdorff距离区域的算法，克服不同手势形状、方向、尺度等对分割的干扰。手势分割实验结果证明提出的算法可以在各种复杂背景及不同光照条件下分割出手势区域，正确率达到99.8%。最后改进了序贯最小优化算法训练二叉树结构的支持向量机多分类器，对实验采集的各种自然条件下九类手势图像的平均识别率超过80%，证明了算法用作普通摄像机下通用人机交互模式的可行性。
手勢識彆技術作為最有前景的一種自然人機交互模式已經成功應用于一些領域。可靠的手勢識彆技術多依賴特定的硬件實現，而這種自然交互模式的普及需要自然環境下基于普通攝像機的通用手勢識彆技術。研究瞭在普通攝像機下對各種複雜揹景、不同光照條件的靜態手勢的分割和識彆技術。首先採用一種鄰域變換算法，剋服不同光照彊度對分割的影響，然後提齣一種求最小平均Hausdorff距離區域的算法，剋服不同手勢形狀、方嚮、呎度等對分割的榦擾。手勢分割實驗結果證明提齣的算法可以在各種複雜揹景及不同光照條件下分割齣手勢區域，正確率達到99.8%。最後改進瞭序貫最小優化算法訓練二扠樹結構的支持嚮量機多分類器，對實驗採集的各種自然條件下九類手勢圖像的平均識彆率超過80%，證明瞭算法用作普通攝像機下通用人機交互模式的可行性。
수세식별기술작위최유전경적일충자연인궤교호모식이경성공응용우일사영역。가고적수세식별기술다의뢰특정적경건실현，이저충자연교호모식적보급수요자연배경하기우보통섭상궤적통용수세식별기술。연구료재보통섭상궤하대각충복잡배경、불동광조조건적정태수세적분할화식별기술。수선채용일충린역변환산법，극복불동광조강도대분할적영향，연후제출일충구최소평균Hausdorff거리구역적산법，극복불동수세형상、방향、척도등대분할적간우。수세분할실험결과증명제출적산법가이재각충복잡배경급불동광조조건하분할출수세구역，정학솔체도99.8%。최후개진료서관최소우화산법훈련이차수결구적지지향량궤다분류기，대실험채집적각충자연조건하구류수세도상적평균식별솔초과80%，증명료산법용작보통섭상궤하통용인궤교호모식적가행성。
As one of the most promising natural Human Computer Interaction(HCI)technology, gesture recognition has successfully applied in several fields. However, most of the dependable methods of gesture recognition depend on specialized hardware. The popularization of the natural HCI technology needs universal gesture recognition method based on universal camera. Therefore, in this paper, the static gesture segmentation and recognition method based on universal camera is researched, and gesture images are captured under different complex background and illuminations. First, an 8-neighbour based transformation algorithm is applied to overcome the impact to the segmentation in different illuminations. Second, an algorithm named Least Average Hausdorff Distance Area(LAHDA)is proposed to overcome the interference of the segmentation in different gesture shape, directions and scales. The simulation results show that this algorithm works well in gesture segmentation under different complex background and illuminations, the correct segmentation rate of the testing samples can be as high as 99.8%. At last, the Sequential Minimal Optimization(SMO)algorithm is improved to train the Binary Tree Support Vector Machine(BT-SVM)multi-classifier, and the average recognition rate of the system is up to 80% on the testing set which is built in this experiment. The experimental results show that the proposed method is a promise method which can be used as a universal HCI mode based on one common camera.
