In the process of automatic maize kernel trait extraction, the parameters of maize kernel can be obtained by machine vision. In the process of machine vision analysis, the accumulation of maize grain affects the extraction of kernel parameters. A method of obtaining information of corn kernel and cyclotron vibration platform was designed. Through the analysis of the stress distribution of accumulation seed cyclotron motion process, the vibration platform swing speed was determined, and the grain accumulation was reduced. Through the vibration kernel accumulation was reduced first, and then maize kernel image was obtained. In the image processing and analysis stage, image of maize kernels was segmented. To solve the segmentation problem of maize kernels, an improved watershed algorithm was proposed. After watershed segmentation, the D-value between the minimum watershed value and the minimum value of the pixel points between adjacent segmentation region were calculated, and then the D-value was compared with T. If the D-value was less than T, the adjacent segmentation region was merged. Segmentation made each maize kernel be a region. The number of maize grains corresponded to the number of regions. After the segmentation was completed, minimum external rectangle of single seed was established by improved Graham scanning method. The length and width of the minimum external rectangle corresponded to the length and width of the kernel. Verification experiments were carried out on the corn seed metering device. The final test results showed that the correct rate of segmentation method proposed on maize grain reached above 98.05% and the determination correlation coefficient between algorithm and manual measured values was above 0.97. The system can realize rapid acquisition of relevant information of maize kernel, and achieve the expected performance index as well as meet practical demands.