CAJ | 학술논문

提出一种改进Snake模型与光流估计相结合的人体运动自动实时跟踪算法.利用角点检测得到接近人体真实轮廓的初始轮廓,减少了迭代次数,降低了Snake模型收敛到局部极值的概率;同时针对Snake模型跟踪不够稳定、容易出现跟踪丢失问题,结合KLT光流法,选取当前帧所得到的轮廓点中的强特征点进行光流估计,将估计结果作为下一帧Snake的初始轮廓,有效地解决了这一难题.实验结果表明改进Snake模型可使初始轮廓形变到人体真实轮廓,同时实现了视频序列中自动、实时的人体跟踪.
제출일충개진Snake모형여광류고계상결합적인체운동자동실시근종산법.이용각점검측득도접근인체진실륜곽적초시륜곽,감소료질대차수,강저료Snake모형수렴도국부겁치적개솔;동시침대Snake모형근종불구은정、용역출현근종주실문제,결합KLT광류법,선취당전정소득도적륜곽점중적강특정점진행광류고계,장고계결과작위하일정Snake적초시륜곽,유효지해결료저일난제.실험결과표명개진Snake모형가사초시륜곽형변도인체진실륜곽,동시실현료시빈서렬중자동、실시적인체근종.
An automatic human motion tracking algorithm combining improved Snake model and optical flow was proposed. The initial contour most close to human body was acquired by corner detection thus decreasing the iterations, and the probability of converging at local extreme value of Snake model was reduced. To resolve the problem that the tracking with Snake model is unstable and often loses the object, the strong feature points were chosen from the contour points in current frame for optical flow estimation, and then the result was chosen as the initial contour of next frame. Experimental results show that improved Snake model can make the initial coutour deform to the actual contour of human body, and realized the automatic and real-time human tracking.