A human-computer interaction system for unmanned aerial vehicle（ UAV） based on human body posture information is designed. Videos captured by UAV are stablizating processed to eliminate effect of UAV vibration on image. Targets in videos are extracted by using foreground information. Convolutional neural network is utilized to classify targets. Human body posture information is identified and UAV will respond according to different recognition results. The result of experiments show that the proposed method can obtain satisfying accuracy of recognition

real-time performance is realized on the airborne embedded computing platform

which shows that it has strong application value.