Multi-strategy dung beetle optimization for robust indoor object detection and tracking for visually impaired people with hybrid deep learning networks

Visually impaired people generally face many troubles in their everyday lives, and technical involvement might help them perform these tasks. Object detection is a significant aspect of computer vision (CV) and machine learning (ML), which plays a substantial part in recognizing and detecting objects in a video or image. Dissimilar objects are detected, and their shape, size, and position data are accurately acquired through object detection. Particularly, visually challenged people are not set to face sudden situations by themselves while moving. Hence, these situations are a significant attack on the protection of the visually impaired. A method is required all over the world to alleviate their day-to-day lives, since it is problematic to help visually impaired individuals in real life. This study proposes a novel Multi-Strategy Dung Beetle Optimization for Robust Object Detection and Tracking with Hybrid Deep Learning Networks (MSDBO-ODTHDLN) model. The proposed MSDBO-ODTHDLN model presents visually challenged people with real-time and reliable object detection and tracking capabilities. Initially, the median filter (MF) is employed in the image pre-processing stage to enhance edge detection and ensure the clarity of objects in diverse environments. Furthermore, the Mask R-CNN method is used for object detection to provide accurate localization and identification of objects within the environment. Moreover, the CapsNet model is the backbone for the feature extraction process to capture spatial hierarchies and intricate patterns. The hybrid method unites a convolutional neural network and a bidirectional long short-term memory (CNN-BiLSTM) model, which is utilized for object detection and classification. Finally, the multi-strategy dung beetle optimization (MSDBO)-based hyperparameter selection process is performed to adjust the analysis outcomes of the CNN-BiLSTM model. To exhibit the improved performance of the proposed MSDBO-ODTHDLN methodology, a comprehensive experimental analysis is conducted under the Indoor object detection dataset. The comparison analysis of the MSDBO-ODTHDLN methodology portrayed a superior accuracy value of 99.77% over existing models.