Optimizing Neural Network to Develop Loitering Detection Scheme for Intelligent Video Surveillance Systems

Using an optimized neural network as machine learning classification model is proposed to develop an enhanced loitering detection scheme for intelligent video surveillance systems. From input surveillance videos, the moving pattern features including the looking-around feature are extracted from the partial trajectories of each pedestrian. For training loitering detection scheme with the extracted features, we tried to optimize the structure and the training parameters of neural network with respect to the detection accuracy and the model building time. And as the experiments, the detection accuracy of the optimized neural network is compared with those of other machine learning classification models such as decision tree, naiveBayesian, and support vector machine. From the experiment results, the loitering detection scheme using the optimized neural network can achieve better detection accuracy of 93.33% than using other models.