Advanced Optimization of Object Detection in Videos: Harmonization of Truncated Average with Dynamic Segmentation by Fuzzy Entropy and Differential Evolution

This paper presents an integrated technique to improve the accuracy of object detection in videos using an intelligent truncated average for background estimation, effectively reducing the influence of transient foreground objects and false detections. A gamma transformation is applied to the residual image before segmentation to enhance the distinction between background variations and foreground regions. The proposed approach combines fuzzy entropy (FE) and differential evolution (DE) for adaptive multilevel segmentation. The fuzzy entropy criterion evaluates candidate fuzzy partitions of the residual histogram, while differential evolution optimizes the fuzzy transition parameters from which the adaptive segmentation thresholds are derived. This approach provides foreground detection, particularly in dynamic video environments. Experimental results on five selected categories of CDnet 2014 achieved an average F-measure of 0.8896. On six selected categories of LASIESTA, the method achieved an average F-measure of 0.8643. These results demonstrate competitive performance compared with several conventional and adaptive background subtraction approaches while maintaining robustness across different video environments. The proposed framework therefore provides a practical basis for moving-object detection in intelligent surveillance and computer vision applications.