Business Hall Abnormal Behavior Detection Based on Cascading Deep Neural Network Model. (October 2019)

• Record Type: Journal Article
• Authors: Yang, He; Wu, Peng; Lin, Guoqiang
• Is Part Of: IOP conference series. Volume 612:issue 3(2019)
• Publication Date: 2019-10
• Subjects: Materials science -- Periodicals; 620.1105
• DOI: 10.1088/1757-899X/612/3/032193
• Languages: English
• ISSNs: 1757-8981

Abstract

Abstract: In the business halls of our country, video cameras are very common, but many videos are analyzed manually, which is time-consuming and labor-intensive, not smart enough. Many of them are used after-the-fact, unable to meet real-time requirements, and the time is relatively tight. However, our method can analyze the video, get real-time analysis results, and perform early warnings. Neuron network is a common method of artificial intelligence for identifying objects in images and videos. This method requires a large amount of sample data to obtain higher accuracy. This paper proposes a behavior detection framework based on cascading depth neural network model. Based on the advanced drfcn neural network, we trained two models and the two models are cascaded. Based on this framework, we implemented a drfcn-service system, which implements the features of drfcn, and flexibly adds plugins to support single model, two-model cascading and even multiple model cascading. Experiments show that the cascading system can significantly improve the accuracy of recognition.