Efficient Aerial Specific Building Recognition Using CRKEM-ORB

Abstract

Detecting a specific building in aerial images presents a significant challenge for aerial package delivery systems, primarily due to the resource-intensive nature of deep learning techniques, which demand extensive training data. The creation of diverse and large-scale datasets for specific building recognition is often cost-prohibitive and time-consuming. In response to these challenges, this paper introduces a new approach that circumvents the need for extensive data collection by harnessing ORB features, while simultaneously reducing redundant keypoints of ORB detection phase through the newly established technique called Clustered Redundant Keypoints Elimination Method (CRKEM). This methodology ensures a satisfactory balance between detection accuracy and execution time for aerial specific building recognition. Our approach offers a viable solution for achieving reliable detection accuracy, even in resource-constrained settings, thanks to the integration with Nvidia's Jetson TX2, a high-performance embedded platform. This capability underscores the potential of CRKEM-ORB in addressing real-world challenges without the burdensome reliance on large datasets.