Towards Efficient Aerial Object Detection: Techniques, Datasets and Challenges

In the past decade, object detection has gained significant attention in computer vision due to variety of its applications including picture retrieval, object tracking, self-driving cars, security, activity recognition, traffic surveillance and many more. Latest research in computer vision is making remarkable progress in multiple directions; aerial object detection is one of the foremost. Object detection in natural scenes has been extensively researched however, challenges remain in aerial object detection due to factors such as variation in scale and shape of objects, occlusion, monotonic appearance, and arbitrary orientation of objects in a scene. Literature highlights deep learning models surpass all others in terms of accuracy and efficiency for aerial object detection and recognition. It is well-known that datasets directly impact deep learning models as deep learning itself is a data driven approach. It is therefore crucial to analyze not only the object detection algorithms in aerial imagery, but also the remote sensing datasets that are available. In this paper, we analyze multiple state-of-the-art models that are used in the field of object detection specific to aerial images. We also present a detailed timeline of the prominent aerial imagery datasets; analyze them in detail and present their strengths and weaknesses. At the end, we discuss various challenges present and future research directions in this field. This research work will further pave way for future advancement in the field of aerial object detection and recognition.