针对传统最小类内方差分割方法计算量大、效率低、单阈值分割、不能多目标分割的缺点，提出了一种改进的基于最小类内方差的蛇形机器人多阈值分割方法．通过提取整幅图像的感兴趣区域( ROI) ，有效减小算法搜索的范围和整体计算量; 根据直方图的多峰值特点，把ROI 区域划分成多个子区域，采用改进的最小类内方差分割法搜索各个局部最优阈值，最终实现蛇形机器人关节组的多阈值分割．实验结果表明，该方法分割效率高，分割效果明显，且在保证实时性的同时提高了目标识别对光线变化的鲁棒性，降低了对步态变化的敏感性．

In order to remove the drawbacks of the traditional single-threshold segmentation methods,namely heavycomputational burden,low efficiency and incapability of multi-target segmentation,an improved multi-thresholdsegmentation method of the snake-like robot is proposed based on the minimum interclass variance ( MIV) .In thismethod,first,the ROI ( Region of Interest) of the whole image is extracted to effectively reduce the search scopeand hence lighten the computational burden.Then,according to the multi-peak characteristics of histograms,the ROI is divided into multiple sub-regions,and the local optimal thresholds of the sub-regions are obtained by meansof the MIV segmentation.Thus,the multi-threshold segmentation of the joint groups of the snake-like robot issuccessfully implemented.Experimental results indicate that the proposed method is of high segmentation efficiency,good segmentation effect,constant real-time performance,stronger target recognition robustness to light intensitychange and lower sensitivity to gait change.