THEORY AND APPLICATIONS OF VIDEO ABNORMAL BEHAVIOR DETECTION
PeiChen Wu, DengBin Xu, LiNing Yuan · Journal of Computer Science and Electrical Engineering · 2024
Video abnormal behavior detection is a research hotspot in the field of computer vision. By extracting the spatiotemporal characteristics of video content, we can determine whether there are abnormal events and their types in the video, and identify the location and time of the abnormal events. Based on supervised/unsupervised learning, this paper systematically combs and summarizes the existing video abnormal behavior detection methods. Starting from the current mainstream modeling idea, the supervision method is described in detail, and the completely unsupervised method is introduced to train the model. The network architectures of different models are compared, and the characteristics of various anomaly detection models in terms of test data sets, usage scenarios, advantages and limitations are summarized. Then, through common evaluation criteria such as frame level standard and pixel level standard, the model is compared and the performance is evaluated. At the same time, the performance of different methods is compared within the class, and the results are analyzed and summarized in depth. Finally, the future development direction is outlined briefly, and the development trend of video anomaly detection from virtual composite dataset, multi-modal large-scale model to lightweight model is discussed.