Enhanced Texture Representation for Moving Targets Classification Using Co-occurrence

摘要

This paper presents a moving targets identification system in more effective computational cost by using Gray Level Co-occurrence Matrix (GLCM) instead of using the other texture descriptors: the conventional LBP histograms and LBPs with co-occurrence matrix. The aim of this work is to develop an enhanced texture analysis based method for the detection and classification of the moving targets in real environment. Firstly, the system distinguished the moving regions from the background regions by using an Adaptive Gaussian Mixture Model (GMM). The gray level (grayscale intensity or Tone) texture features on a co-occurrence matrix will be extracted from each segmented moving block by the four texture features, energy, homogeneity, correlation and contrast in four directions (0 , 45 , 90 , and 135 ) and quantized into a feature vector. These exploited texture features will be used to classify the moving objects using the Support Vector Machine (SVM) classification learner. The walking-dog-14-0-3 test sequence from UCF11 dataset is used in experimentation to show the effectiveness of the proposed feature method.