Through the big data video retrieval technology, and the deep learning-based target detection algorithm, it can, on a terminal, acquire and recognize targets such as pedestrians, vehicles, and items-left in complicated scenarios in a self-adaptive way. Meanwhile, because of the video abstract retrieval, mining and analysis technologies employed by the supercomputer server on the background, it is highly sensitive to the special people or abnormal behaviors in the detected scenarios, including pedestrian attribute recognition, knife or pistol recognition, recognition of abnormal behaviors such as climbing, recognition of illegal acts such as flocking together illegally, killing a person with a knife, shooting with a pistol, etc.