在危险作业场所中，采用机器视觉技术对进入工作现场人员进行实时统计。文章首先使用颜色特征并结合区 域生长的方法对人的头部区域进行初步定位，接着基于圆轮廓对人头区域实现确定。在提取连续帧人头区域的基础上进行目 标跟踪，建立预测函数辅助人头运动轨迹的获取，最终根据运动轨迹线判断目标是否进入或离开生产作业场所。实验表明，本 文提出的方法准确率在90%以上，具有良好的实用性与学术价值，对于确保安全生产具有重大现实意义。

The paper proposes a statistical method of headcount based on machine vision technology, which is implemented in quota management in dangerous production place. The head’s region is extracted roughly with hair color feature and regional growing algorithm. And then, the head region is matched by the mean of circle contour. Obtaining a series of continuous frame images, target tracking can be realized. The detail is that moving forecasting function is set assisting to obtain head's moving trajectory. So whether the people enter or leave production place can be judged according to trajectory. Finally, the experiment proves the practical effectiveness and academic value of the method, the accuracy can reach 90% above. It is also very significant for safety in production.