Survey on Object Detection in Aerial Imagery

Abstract

Object detection is a process of locating and identifying entities in an image or video. It is a fundamental process in many real-life applications. It can be used for aerial images to detect different objects, which can be used for tasks like surveillance and analysis. Being a crucial process in these applications it is also a challenging problem. It involves challenges like obscured objects, cluttered backgrounds, unpredictable orientations, and comparatively small-scale objects for detection. Along with this aerial image gives overhead view of objects which makes feature extraction difficult and can cause false alarm. In practice, an object and classification detection model are often required to perform well in terms of speed and detection accuracy. Although many traditional detection models have shown a decent performance result by using the multiple templates in a sliding-window manner and imagery pyramid, these methods are still inefficient. Satellite images contain vast information for processing and analysis, but the human eye is not able to recognize the little changes like intensity color and texture. To extract information from aerial images, a segmentation method could be applied which is a difficult and important task in image analysis. This can be primarily done through edge detection, wavelet-based method, and multispectral classification. The customary detection techniques have made remarkable progress with horizontal bounding boxes (HBBs) because of CNNs. Nevertheless, HBB detection techniques still have some limitations that include missed detection and redundant detection regions, especially for densely clustered and strip shaped objects. This paper gives a detailed study and discussions on the idea of existing techniques for detection of objects in aerial images.