Detecting and identifying occluded and camouflaged objects in low-illumination environments

Abstract

One of the prevailing areas of contemporary research involves the differentiation and identification of diverse objects within a given scene through automated systems. The field of study under consideration presents a multitude of obstacles, including but not limited to issues such as diminished lighting conditions, occlusion, and camouflage. The captured image exhibits variations in illumination, resulting in uneven brightness, reduced contrast, and the presence of noise. The fundamental basis of computer vision algorithms lies in the process of extracting features from datasets and subsequently discerning these features through neural networks. The task of extracting distinct feature key points from images captured under low lighting conditions is exceedingly challenging. To address this issue, the present study seeks to employ deep learning models to implement image enhancement techniques specifically designed for low-light conditions. The primary emphasis lies in obtaining key feature points that are differentiable, thereby enabling the utilization of this annotated data for specific tasks such as object detection. The task of identifying occluded and camouflaged objects has been successfully accomplished, yielding an impressive accuracy rate of 93% in total. The mean average precision has been achieved as 85% which is reasonably high compared to many earlier works.