Real-Time Object Detection for 360-Degree Panoramic Image Using CNN

Abstract

The object detection for 360-degree panoramic images is widely applied in many areas such as automatic driving, navigation of drones and driving assistance. Most of state-of-the-art approaches for detecting objects in ordinary images cannot work well on the object detection for 360-degree panoramic images. As a 360-degree panoramic image can be considered as a 2D image which is the result of a 360-degree panoramic sphere being expanded along the longitude line, objects in it will be twisted or divided apart and the detection will be more difficult. In this paper, we present a real-time object detection system for 360-degree panoramic images using convolutional neural network (CNN). We adopt a CNN-based detection framework for object detection with a post-processing stage to fine-tune the result. Additionally, we propose a novel method to reuse those exisiting datasets of ordinary images, e.g., the ImageNet and PASCAL VOC, in the object detection for 360-degree panoramic images. We will demonstrate with several examples that our method yields higher accuracy and recall rate than traditional methods in object detection for 360-degree panoramic images.