CAJ | 학술논문

提出一种基于星形距离轮廓特征和LDCRF模型的在线行为识别方法。对视频中已分割出的人物姿态提取轮廓，求取人体轮廓质心及其到轮廓采样点的星形距离向量，以该向量参数化人体运动姿态特征，对原始姿态特征向量进行小波变换，降维的同时获得姿态的多分辨细节信息。利用潜动态条件随机场模型（latent‐dynamic conditional random ， LDCRF）对人体行为特征建模，进行在线识别。比对CRF、HCRF、LDCRF模型对10种不同行为的识别结果，对比结果表明，相比CRF和HCRF ，该模型对连续行为序列有较强的识别能力，具有更好的稳定性。
제출일충기우성형거리륜곽특정화LDCRF모형적재선행위식별방법。대시빈중이분할출적인물자태제취륜곽，구취인체륜곽질심급기도륜곽채양점적성형거리향량，이해향량삼수화인체운동자태특정，대원시자태특정향량진행소파변환，강유적동시획득자태적다분변세절신식。이용잠동태조건수궤장모형（latent‐dynamic conditional random ， LDCRF）대인체행위특정건모，진행재선식별。비대CRF、HCRF、LDCRF모형대10충불동행위적식별결과，대비결과표명，상비CRF화HCRF ，해모형대련속행위서렬유교강적식별능력，구유경호적은정성。
The online behavior recognition based on star distance and LDCRF model was presented .The contour for the segmen‐tation of character gesture in the video was extracted ,and the human movement characteristics were parameterized through the calculated star distance between body contours centroid and sampling point ,then its dimensionality was reduced using wavelet transform .A model for human behavior characteristics was built by using LDCRF and the behavior was recognized .During the simulation experiment ,by comparing the identification results of CRF ,HCRF and LDCRF of ten different behaviors respective‐ly ,the model LDCRF is verified to be better in recognizing continuous behavior sequence than CRF and HCRF .Moreover ,it has better stability .