Detection of ground holes using Deep Learning for Surveillance

For border area security and surveillance, maintaining vigilance and detecting potential threats is of utmost importance. While surveillance drones have proven effective in enhancing border area monitoring, there are instances where ground pits can raise suspicion. Ground pits, often excavated or dug into the earth, can serve as hidden locations for various illicit activities. The inconspicuous nature of ground pits makes intruders attractive to criminals attempting to evade detection. Deep learning has shown promising potential in automating object detection through visual data. Integrating the deep learning model into drones would provide a more comprehensive and robust surveillance system. In this paper, an image dataset of ground pits referred to as Ground Pit Image Dataset (GPID) is developed to train and test YOLO. This dataset contains 300 images of different ground pits on various surfaces, captured through drones and annotated using online tools. YOLO has provided more than 90% accuracy, which is better than other deep-learning models.