Abstract: In order to ensure the safety of plant protection unmanned aerial vehicles (UAVs), it was required to have the ability of the automatic obstacle avoidance. So a new obstacle detection method based on structured light vision was proposed. In order to improve the real-time performance of obstacle detection, based on the analysis of the latest general processor unit (GPU), the obstacle detection system of plant protection UAVs based on embedded platform was studied mainly. By mapping the obstacle image processing algorithm to GPU hardware resources to complete the parallel computation, the efficiency of the algorithm was greatly improved. Experiments show that the obstacle detection system based on embedded CPU-GPU achieves a speedup rate of about 46.15 by comparing the processing algorithms of CPU and CPU-GPU on the premise of ensuring the complete outline of the obstacle, and time-consuming of the acquisition and processing is about 48.985 ms. The system has the advantages of obvious effect of processing and good real-time performance, which lays the foundation for real-time obstacle detection and further obstacle avoidance of plant protection UAVs.