Abstract: 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 two-dimensional 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 4.85 s. 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.