视频异常行为的检测对保障公共安全至关重要,对基于深度学习的异常行为检测算法进行了分类与总结。首先,介绍了异常行为检测的整体流程。然后,根据神经网络训练的方式,从有监督学习、弱监督学习和无监督学习三个方面论述了深度学习在异常行为检测领域的发展与应用,同时分析了不同训练方式的优缺点。最后,介绍了常用数据集以及性能评估准则,分析了不同算法的性能,并展望了未来的发展方向。

The detection of video abnormal behavior is paramount to ensure public safety. In this paper, the abnormal behavior detection algorithm based on deep learning is classified and summarized. First, the overall process of abnormal behavior detection is presented. Then, based on the neural network training method, the development and application of deep learning in the field of abnormal behavior detection are discussed from three aspects: supervised learning, weakly supervised learning, and unsupervised learning, and the advantages and disadvantages of different training methods are analyzed. Finally, commonly used datasets and performance evaluation criteria are presented, the performance of the different algorithms is analyzed, and future directions are discussed.