基于Stereo Camera-YOLOv5自然环境下百香果检测与定位模型

Detection and localization model of passion fruit in natural environment based on Stereo Camera-YOLOv5

针对百香果采摘机器人在自然环境中作业时受复杂光线及遮挡影响,难以快速精确地检测及定位成熟百香果的问题,提出一种基于Stereo Camera-YOLOv5自然环境下成熟百香果检测及定位模型。针对自然环境下光线以及遮挡的影响,通过MSRCP算法、随机遮挡、数据增扩等图像处理算法对原始数据集进行优化。将优化的数据集输入到YOLOv5网络中训练出最优模型,在检测代码中嵌入双目立体视觉算法。该模型对自然环境下百香果进行检测及成熟度判断,将判断为成熟的百香果进行图像处理,并提取到中心点二维坐标。通过立体匹配及视差计算得到中心点的三维坐标。田间试验结果表明,该模型的目标检测准确率为97.8%,总体准确率为90.2%,平均运行时间为4.85 s。该系统鲁棒性强、实时性好,能够更好地实现自然环境下成熟百香果的检测及定位,为百香果采摘机器人后续工作奠定基础。

Aiming at the problem that the passion fruit picking robot is affected by complex light and occlusion when it operates in the natural environment,and it is difficult to quickly and accurately detect and locate the ripe passion fruit model,a detection and location model of ripe passion fruit in natural environment based on Stereo Camera-YOLOv5 based is proposed.Firstly,aiming at the influence of light and occlusion in the natural environment,the original data set is optimized through image processing algorithms such as MSRCP algorithm,random occlusion and data augmentation.The optimized data set is put into the YOLOv5 network to train the optimal model,and the binocular stereo vision algorithm is embedded in the detection code.The model detects and judges the maturity of passion fruit in the natural environment,processes the image of the passion fruit judged to be ripe,and extracts the twodimensional coordinates of the center point.The three-dimensional coordinates of the center point are obtained through stereo matching and parallax calculation.The field test results show that the target detection accuracy of the model is 97.8%,the overall accuracy rate is 90.2%,and the average running time is 4.85s.The system has strong robustness and good real-time performance,and can better realize the detection and positioning of ripe passion fruit in the natural environment,laying a foundation for the follow-up work of the passion fruit picking robot.