基于单目视觉的移栽钵苗叶片朝向的调整方法

Monocular vision-based method for direction adjustment of transplanting potted-seedling leaves

为解决钵苗移栽时叶片朝向一致的问题,研究单目视觉钵苗叶片的调向方法。首先,均隔8°提取钵苗旋转1圈视频中的45帧图像作为目标对象,对每帧图像依次采用灰度变换、阈值分割和细化等算法进行图像处理后,将8邻域中只有一个与其灰度值相同的中心像素判定为叶片末梢尖点(兴趣点)。然后,通过平滑相似度函数将兴趣点路径跟踪问题简化为对前一帧兴趣点的临近位置均值移位搜索的最大相似度;根据跟踪对象运动矢量方向和速度的连贯性,通过最小化函数从临近位置内存在的多个兴趣点中筛选出目标兴趣点;采用最小化接近一致性代价幻影点函数产生一个帧间幻影点代替因叶片相互遮挡可能丢失的兴趣点。依次根据各帧图像的多个叶片兴趣点获得其相应跟踪路径和对应方位。最后,将叶片兴趣点中存在最大横坐标值的一对对称兴趣点作为目标,当其中一个兴趣点处在最大横坐标位置时,根据另一个兴趣点的当前位置确定钵苗顶杆的调整转角θ/2,将叶片朝向调整到与横轴平行后完成调向。试验结果显示,50株钵苗的最小、最大调向偏差值分别为1.51°和25.31°,94%的调向偏差值小于15°。钵苗叶片调向方法可以满足钵苗移栽机的叶片调向要求。

In order to avoid mutually sheltered between potted-seedlings leaves and improve the light receiving environment of potted-seedlings’ photosynthesis, the orientation of potted-seedlings leaves need to be adjusted to the same direction when potted-seedlings are transplanting. The orientation adjustment method of potted-seedlings leaves was presented using monocular vision while the perpendicularity and height detection of potted-seedlings had been completed using monocular vision at the early stage of the research work. First of all, the video of potted-seedlings rotating 360° was captured by an industrial camera on its side. 45 key frame images were extracted from the video at intervals of 8°, and were used as the target object. Each of extracted key frame images were processed by an 8-bit image gray-scale transformation, and image threshold segmentation of threshold 114, and image thinning. If only one pixel gray value was identical to the center pixel gray value in an 8 neighborhood of center pixels, the center pixel was determined to be the potted-seedling leaves’ ending point （interest points） after their image was processed. Then, the path tracking problem of interest points were simplified to a near location shifting maximum similarity search problem using a smooth similarity function. If there are multiple interest points targeted in a close location, tracking objects were filtered through the coherence of the motion vector direction and speed using a minimizing function. If the displacement of the same interest point between two frames was larger than the preset value, the phantom points were created instead of lost interest points by a minimized near coherence phantom point function. The potted-seedling leaves’ track path and corresponding directions of interest points were obtained according to the method. Finally, it was used as the target object for a pair of interest points existing x coordinate maximum values in a camera imaging plane during the course of potted-seedlings rotating 360°. The adjustment rotation angleθof potted-seedling leaves was determined by the current location of another interest point using an inverse trigonometric formula, when one of interest points was at the maximum x coordinate position. A potted-seedling ejector was driven to revolveθ/2 by electric motor, and the orientation of the potted-seedlings leaves were adjusted the same direction with x, and the adjustment operation of the potted-seedling leaves was completed. Experiment results showed that minimum leaves adjustment direction deviation of 50 potted-seedlings was 1.51°, and the maximum deviation was 25.31°, and the percentage of the adjustment direction deviation less than 15° were 94%. Therefore, this method can meet the requirements of seedlings leaves’ adjustment direction for potted-seedling transplanting.