A Systematic Review of Abnormal Behavior Detection: Convolution Neural Networks and Long Short Term Memory Models

Abnormal behavior detection is interesting topic computer vision and this is required in surveillance system. Various researches were carried out in abnormal behavior detection using machine learning methods. In this paper, the recent papers in abnormal behavior detection is reviewer with advantages and limitations. Many researches in abnormal behavior detection is based on the Convolutional Neural Network (CNN) for feature extraction and classification. Some of the researchers were involves in applying Long Short Term Memory (LSTM) model for the abnormal behavior detection. Autoencoder models were applied for spatiotemporal analysis of abnormal behavior detection. The CNN models have limitations of overfitting problem and imbalance data problem. The LSTM models have vanishing gradient problem that affects the performance of the model. The hybrid model of CNN and LSTM have high computational complexity and requires high resources for computation. The review shows that VGG-19 model have higher performance in abnormal behavior classification compared to other CNN models. An efficient model is required for abnormal behavior detection to overcome overfitting and vanishing gradient problem.