摘要
针对采用双目视觉测距进行无人机实时巡线测距时计算复杂、实时性不高的问题,提出了结合改进的SIFT算法的双目视觉测距方案.该方案采用类单目的双目视觉模型,以减少计算变量;采用缩小尺度空间个数、降低特征向量维度、街区距离代替欧式距离的方法对传统的SIFT算法进行改进,以提高两摄像机同时拍摄的两幅图片同一特征点的匹配效率.结果表明:该方法匹配时间缩短了32%,测量最大误差率降至3. 75%,可满足无人机快速实时巡线测距的精度要求.
In order to solve the problem of complex calculation and low real-time performance when using binocular vision ranging for UAV real-time line patrol, a binocular vision ranging scheme combining improved SIFT algorithm was proposed. A binocular vision model similar to monocular vision was adopted to reduce computation variables, the traditional SIFT algorithm was improved by reducing the number of scale spaces, reducing the dimension of feature vectors, and replacing the Euclidean distance with the block distance, so as to improve the matching efficiency of the uniform feature points of two images taken simuhaneously by two cameras. The results showed that the matching time was reduced by 32% and the maximum error rate was reduced to 3.75%. The method could meet the accuracy requirements of UAV fast real-time line patrol ranging.
作者
陈建明
时铭慧
CHEN Jianming;SHI Minghui(College of Electric Poewr,North China University of Water Resources and Electric Power,Zhengzhou 450045,China)
出处
《轻工学报》
CAS
2018年第5期90-96,共7页
Journal of Light Industry
