Enhancing Object Detection Using Information Theory and Network Coding

This research explores the challenge of dynamic environment object detection. Changing environment factors adversely affect the feature detection and robustness of the model. The approach proposed homogenizes the principles from Information Theory and Network Coding to enhance the overall final matching accuracy and robustness of the model. The effectiveness is improved by systematically using the concepts of entropy, mutual information, Principal Component Analysis, Random Linear Network Coding and Support Vector Machines. The developed framework maintains distinctive feature recognition and classification performance with rapid environment changes. The proposed and the baseline model is run in python using OpenCV for obtaining the experimental result. By including these robust feature extraction techniques along with Scale-Invariant Feature Transform, in dynamic environments, the model achieves a final matching accuracy of 72.4% and an F1-score of 76.1%. In contrast, the baseline model achieves a final matching accuracy of 43.8% and an F1-score of 44%. The improvement in performance shows the model's enhanced resilience and robustness with the changing environmental conditions, contributing valuable prospects for real-time applications where environmental variability is a factor to be considered.