Cross-camera Tracking Technology for Intelligent Communities Based on Deep Learning and Fuzzy Clustering Algorithms

Abstract

In the current construction of smart communities, continuous tracking across cameras has become the key to improving security efficiency. This study aims at the existing problems such as identity consistency maintenance, noise interference and perspective changes in cross-camera target tracking in intelligent communities. Furthermore, it proposes an intelligent cross-camera tracking technology based on an improved fuzzy C-means clustering algorithm and deep learning. In the image segmentation stage, this research improves the robustness of the segmentation algorithm in complex monitoring scenarios by constructing an affinity matrix that fuses pixel and regional information, and introducing spatial regularization terms and fixed clustering center strategies. In the tracking stage, this research combines large-kernel convolutional attention and masked autoencoder self-supervised learning to achieve discriminative feature extraction and efficient cross-camera matching, and improve the accuracy of cross-camera tracking. According to experimental results, the suggested improved fuzzy C-means clustering algorithm achieved segmentation accuracy of 90.5% and 83.8% on two public datasets, and the mean intersection over union under 30% speckle noise was 94.8% and 93.7%, respectively. In the actual monitoring scenario, the proposed cross-camera tracking method achieved an identity preservation rate of 94.7%, only 38 identity switches, and an average tracking delay of only 65.6ms. The findings demonstrate that the suggested approach offers a workable technical solution for intelligent community security systems and can significantly increase the precision and usefulness of cross-camera tracking.