Abnormality Retrieval Method of Laboratory Surveillance Video Based on Deep Automatic Encoder

Abstract

Aiming at the problem that abnormal behavior is difficult to distinguish from normal behavior, a retrieval method for abnormal behavior of laboratory security surveillance video based on deep automatic encoder is proposed. Firstly, the fuzzy median filtering algorithm is used to reduce the noise of the collected laboratory security surveillance video, and then the YUV spatial chromaticity difference method is used to divide the foreground and background of the video, and the illumination degree in the video is determined. The diagonal model and codebook clustering idea are used to compensate for global and local lighting mutations. Finally, the preprocessed video is input into the mixture model, which is based on the deep automatic encoder and combined with the Gaussian mixture model, and the abnormal behavior retrieval results are output. The experimental results show that the proposed method has good security surveillance video preprocessing effect, large AUC, small error rate of abnormal behavior retrieval, and high operation efficiency.