摘要
从双目定位的几何分析出发,提出了一种基于断层切片透视的双目测量模型.该模型首先建立了不同深度距离下物点视差与深度距离的非线性关系,然后在物点深度距离已知的情况下,使用预先建立的各个断层切片距离下的神经网络映射模型,根据物点的图像坐标确定其2维X,Y坐标.当切片距离与建模时深度距离不一致时,采用最近邻原则到最接近距离下的模型进行近似.结果表明:该测量模型从深度距离223mm到1183mm的97个断层切片位置,在Z方向的最大绝对误差不超过4.5mm,在X,Y方向的最大绝对误差均不超过2.0mm,偏离真实点空间距离的误差在0.2416~4.8590mm之间变化.试验表明:该模型可以满足果实空间定位的测量精度要求.
According to geometric analysis of binocular location, a kind of binocular measurement model based on fault slice perspective was presented in this paper. Nonlinear relation between optical parallax of object point and different depth distance was firstly established. Neural network model was established in advance on each slice plane to determine planar X and Y coordinate of object point according to known depth distance. When slice distance was different from modelling depth distance, nearest neighbor rule was adopt to model under proximate distance. The results show that maximum absolute error of model in depth direction is less than 4.5 mm, and those in X and Y directions are less than 2.0 mm. The error of spatial distance with real object point varies from 0. 241 6 mm to 4. 859 0 mm when 97 fault slice positions change from 223 mm to 1 183 mm. The tests of locating tomato show that the model can meet the requirement for precise locating fruit.
出处
《江苏大学学报(自然科学版)》
EI
CAS
北大核心
2011年第3期249-253,共5页
Journal of Jiangsu University:Natural Science Edition
基金
国家自然科学基金资助项目(50805067)
中国博士后基金资助项目(20080441073)
江苏大学高级专业人才科研启动基金资助项目(07JDG078)
关键词
双目测量
神经网络
定位
番茄
模型
binocular measurement
neural network
location
tomato
model