Multichannel Correlation Clustering Target Detection

Abstract

During target tracking, certain multi-modal background scenes are unsuitable for off-line training model. To solve this problem, based on the Gaussian mixture model and considering the pixels’ time correlation, a method that combines the random sampling operator and neighborhood space propagation theory is proposed to simplify the model update process. To accelerate the model convergence, the observation vector is constructed in the time dimension by optimizing the model parameters. Finally, a three channel-multimodal background model fusing the HSI color space and gradient information is established in this study. Hence the detection of moving targets in a complicated environment is achieved. Experiments indicate that the algorithm has good detection performance when inhibiting ghosts, dynamic background, and shade; thus, the execution efficiency can meet the needs of real-time computing.