Object Detection on the ExDark Dataset using image pre-processing and different object detection algorithms Detecting objects present in low-lighting conditions has always been a challenging task. Reviewing enhancement techniques for low-light object detection is crucial in research applications like surveillance and automated driving. The first task is to choose a benchmark dataset unlike other datasets with a low number of images in dark lighting. The ExDark dataset is one such dataset that exclusively contains instances in low-light. Equisampling followed by comparing the results using cutting-edge object detection pipelines with enhancement techniques is the primary task of the present work. In this paper, we analyze the methodologies that led to the improvement in the performance of the deep learning-based object detection algorithms. The methodology is based on the enhancement of the input images with low light intensity before performing object detection. We also formulated a new method to obtain better bounding boxes predictions for object detection using RCNN.
Model Weights available at: https://drive.google.com/drive/folders/1tRt-JVO2icQFF8JJng3XlWcacRU7POEn?usp=sharing
Authors: Arrun Sivasubramanian, Prashanth VR, Arun Prakash J, Dharshan Kumar KS, Sowmya V, Sajith Varrier VV Link to paper: https://link.springer.com/chapter/10.1007/978-981-99-3656-4_27