CAJ | 학술논문

基于视频流的运动人体行为识别是一项既具有挑战性同时又非常具有广阔应用前景的研究课题。行为识别是基于人体目标识别和人体跟踪更高级的计算机视觉部分，研究出一种健壮的行为识别算法具有重要的理论意义和广泛的应用前景。利用在视频的基础上提取出位置分布图、大小分布图等一系列的属性将人的行为进行分类。采用基于帧间差分和改进混合高斯模型的运动人体分割算法，解决了复杂背景下的运动目标检测问题。实验数据对提出的新的行为描述方法进行了各种指标的讨论，验证了本文提出的算法的合理性与高效性。
기우시빈류적운동인체행위식별시일항기구유도전성동시우비상구유엄활응용전경적연구과제。행위식별시기우인체목표식별화인체근종경고급적계산궤시각부분，연구출일충건장적행위식별산법구유중요적이론의의화엄범적응용전경。이용재시빈적기출상제취출위치분포도、대소분포도등일계렬적속성장인적행위진행분류。채용기우정간차분화개진혼합고사모형적운동인체분할산법，해결료복잡배경하적운동목표검측문제。실험수거대제출적신적행위묘술방법진행료각충지표적토론，험증료본문제출적산법적합이성여고효성。
Moving human behavior recognition based on a video stream is a challenging research topic, which has broad application prospects at the same time. Behavior recognition is a part of more advanced computer vision based on the human target recognition and tracking human body. It has important theoretical significance and broad prospect to develop a robust behavior recognition algorithm. On the basis of the video, using a series of properties to classify the human behavior, location distribution, size distribution, etc. Human Movement based on the frame difference and improved Gaussian mixture model segmentation algorithm to solve the problem of moving target detection under complex background. The experimental data discussed various indicators on the behavior of the proposed new description method, which verifies the rationality and efficiency of the proposed algorithm.