刘刚教授团队：基于深度图像的多姿态肉牛体尺自动测量方法（《智慧农业（中英文）》2022年第4期）

文章图片

图1　肉牛数据采集装置

Fig. 1  Data acquisition device of beef cattle

图2　相机矫正前后肉牛数据采集效果

Fig. 2  Data collection effect of beef cattle before and after depth camera correction

图 3　肉牛深度图像预处理结果

Fig.3  Deep image preprocessing results of beef cattle

图4　利用Zhang-Suen算法获取肉牛主骨架

Fig. 4  Getting the main skeleton of beef cattle by using Zhang-Suen algorithm

图5　凸包分析方法获取肉牛图像上下轮廓边缘

Fig. 5  Getting the upper and lower contour edges by convex hull analysis method

图6　欧氏距离确定肉牛头部去除点

Fig. 6  Determining the head removal point by Euclidean distance

图7　多姿态下肉牛头部去除图像效果

Fig. 7  Head removal effect of beef cattle in multiple poses

图8　P_i点U弦长曲率的支持邻域

Fig. 8  Support neighborhood of U chord curvature at point P_i

图9　肉牛图像毛刺点类型

Fig. 9  Burr point types of beef cattle image

图10　划分5×5的卷积核正负区域

Fig. 10  Divide the positive and negative areas of the convolution kernel of 5×5

图11　肉牛体尺计算示意图

Fig. 11  Schematic diagram of beef cattle body size calculation

图12　肉牛姿态不端正度评价θ角度计算方法

Fig. 12  Evaluation method of beef cattle posture

skew degree θ 

图 13　在不同低头幅度H_u和姿态歪斜程度θ下肉牛头部去除结果

Fig.13  Head removal results of beef cattle under different head-down posture H_u and posture skew degree θ

图14　U弦长曲率算法改进前后计算得到肉牛轮廓曲率曲线

Fig. 14  Curvature curve of beef cattle contour calculated by U-chord length curvature algorithm before and after improvement