Anomaly Detection Algorithm Based on Deep Learning and Gaussian Mixture Model
Current approaches for anomaly detection cannot tackle both human-related abnormal behavior recognition and non-human anomaly recognition.Moreover,the detection accuracy is low especially in the case of occlusion.To solve this problem,the network based on GRU is proposed to train the structural feature segments of the human body posture in the normal frame.In do-ing so,this approach is able to anti-occlusion.This paper further introduces an improved Gaussian mixture model with adaptive K value to extract abnormal situations(such as cars,bicycles,remnants,etc.)that are not related to human.Experiments show that the anomaly detection AUC of proposed method increased by 0.025 and 0.034 on Avenue dataset and ShanghaiTech dataset respec-tively.
国家科技期刊平台
NSTL
万方数据
