基于支持向量机的无人机视觉障碍检测

Vision Obstacle Detection for UAV Based on Support Vector Machine

自主障碍检测与回避是无人机低高度飞行时保障其生存性的一项关键技术,有重要的研究意义。通过对机器视觉原理的研究,考虑到支持向量机方法能同时减小匹配难度和计算量,实时性能、泛化性能良好,故采用该方法通过离线监督学习,将无人机前视图像分割为天空与非天空2部分,并将非天空部分作为需要回避的障碍,实现无人机基于视觉的障碍检测系统,为后续的视觉制导提供信息。实验结果表明,支持向量机能有效准确地实现图像的天空分割,并具有良好的泛化性能。

The autonomous obstacle detection and avoidance of unmanned aerial vehicle（UAV） in low-level flight is a key technology to ensure its survivability.Therefore,it is important to develop the technology.After the study of the computer vision principle and off-line supervised learning,since the consideration that the support vector machine（SVM） has the properties in generalization and real-time performance,and can reduce the matching difficulty and computation quantity at the same time,SVM is used to segment a forward-looking image into two parts（sky and non-sky）,in which the non-sky part is considered as the obstacle that UAV should avoid.A vision-based obstacle detection system was achieved for UAVs and the information for subsequent visual navigation was provide.The experimental results indicate that SVM can segment forward-looking images accurately and has good generalization performance.