Automated Object Rescue From Open Borewells using Enhanced RCNN Framework
Pokala Sai Sparsha Prasad, Praveen Kumar Gupta · 2025
In rural and urban settings as well as frequently result in mishaps involving people or animals caught inside, open borewells seriously compromise public safety. The limited and deep structure of rescue efforts makes them more difficult since traditional methods take time and are prone to errors. Detecting and rescuing objects caught in bore wells is made difficult by low light, cluttered surroundings, and the necessity of accurate localization and classification of the trapped entities. Conventions lack real-time monitoring and the degree of accuracy required for rapid rescue operations. This work attempts to automate the detection and rescue process by means of a deep learning based solution leveraging Region-Based Convolutional Neural Networks (RCNN). Apart from real-time monitoring capability, the RCNN structure combines several object detection and classification modules. To address typical challenges including low-light conditions, occlusions, and a range of object shapes, the model is trained on a varied dataset. The system can achieve good feature extracting and accurate localization by means of transfer learning. The results of the experimental evaluations show that the proposed system detects and classifies objects caught, even in demanding conditions, with very high degree of accuracy. The RCNN-based architecture significantly improves the response times for rescue operations by means of an accuracy of over 95% in object localization. Testing it under several borewell conditions confirmed its scalability and dependability, so verifying its practical use for rescue operations.