改进的Otsu算法及在运动目标检测中的应用研究

针对Otsu自适应阈值分割算法中阈值搜索精准度较低、效率不高的问题，本文结合视频图像序列的帧间相关性，利用遗传算法的全局寻优优势，及模拟退火算法较好的爬山性能，提出一种改进的Otsu算法，并与三帧差分法相结合应用于视频运动目标检测。实验证明，该算法相对Otsu算法和对比算法减少阈值分割中的寻优尝试次数，使最优阈值的选取更精确，并提高了目标检测效果。

Research on an improved Otsu algorithm and its application in moving objection detection

Considering the problem of the Otsu adaptive threshold segmentation algorithm which the accuracy of the threshold search is lower and the efficiency is not high in threshold search precision, in this paper, we combined with the correlation between frames of video sequences, used genetic algorithm with the advantage of global searching, and simulated annealing algorithm with the better performance of mountain climbing, to improve Otsu algorithm. Then we contact with three frame difference method to detect moving objection in video sequences. Experimental results show that, the method reduced segmentation threshold optimization attempts, selected more accurate threshold compared with the traditional Otsu algorithm and comparison algorithm, and improved the target detection effects.