Improved SSD network for Oriented Object Detection

In recent years, object detection techniques based on convolutional neural network (CNN) have achieved many breakthroughs and have been widely used in various object detection problems. However, rectangular bounding box representations are not suitable for ubiquitous directional objects. This study proposes a series of improvements to the SSD model to achieve multi-directional object detection capabilities. Improved SSD directly returns all vertices of an object instead of rotating or sliding vertices on the bounding rectangle. In the experiment, we combined unmanned vehicle and fuzzy controller to verify the effect of parking space detection.