The Detecting And Tracking Algorithm For Multi-objects Based On RBFUN Of L-GEM

Moving object detecting and tracking play an important role in many areas such as intelligent video surveillance. In recent years, a large number of algorithms related to detecting and tracking have been proposed. However, during to the diverse circumstances of the background and the large calculation of algorithms, it’s difficult to meet the needs of the real-time detecting and tracking in the actual application.For the target of real-time detecting and accuracy tracking, this paper presents algorithms based on the combination of two main algorithms framework and proposes following improved algorithmUse the modified Gaussian Mixture Models to complete the extraction of motion area. Taking into account the Gaussian Mixture Models will be deteriorated in the situation of light mutations, this paper proposes the modified algorithm by adjusting the learning efficiency of Gaussian Mixture Models when light changes suddenly, the algorithm will weaken the light sensitivity of the model to make the model more adaptable.Use the RBFNN based on L-GEM to complete the classification of moving object. The RBFNN based on L-GEM, because of its good generalization ability and fast convergence rate, has been widely used in many applications, but it have not been achieved in this application so far.Use the resampling based on stochastic sensitivity to balance the weight of the samples. For frame images, the number of target areas in the images often far less than the background areas, and the misclassification will take great cost. In the current algorithm, many classifiers such as SVM did not take this into consideration. In this paper, we use the resampling based on stochastic sensitivity to balance the weight of the samples to improve the training performance of the classifier.Based on these three aspects above, this paper improves the current algorithm, and the results of experiment demonstrate that the algorithm ensure not only the need of accuracy of detecting and tracking, but also real-time.