Skoči na glavni sadržaj

Izvorni znanstveni članak

https://doi.org/10.17559/TV-20171229024444

Applying the Convolutional Neural Network Deep Learning Technology to Behavioural Recognition in Intelligent Video

Lele Qin orcid id orcid.org/0000-0003-3222-6453 ; School of Economic Management of Hebei University of Science & Technology, Shijiazhuang, Hebei, 050000, China
Naiwen Yu ; Polytechnic College of Hebei University of Science & Technology, Shijiazhuang, Hebei, 050000, China
Donghui Zhao ; School of Information Science and Engineering of Hebei University of Science & Technology, Shijiazhuang, Hebei, 050000, China


Puni tekst: engleski pdf 1.803 Kb

str. 528-535

preuzimanja: 1.096

citiraj


Sažetak

In order to improve the accuracy and real-time performance of abnormal behaviour identification in massive video monitoring data, the authors design intelligent video technology based on convolutional neural network deep learning and apply it to the smart city on the basis of summarizing video development technology. First, the technical framework of intelligent video monitoring algorithm is divided into bottom (object detection), middle (object identification) and high (behaviour analysis) layers. The object detection based on background modelling is applied to routine real-time detection and early warning. The object detection based on object modelling is applied to after-event data query and retrieval. The related optical flow algorithms are used to achieve the identification and detection of abnormal behaviours. In order to improve the accuracy, effectiveness and intelligence of identification, the deep learning technology based on convolutional neural network is applied to enhance the learning and identification ability of learning machine and realize the real-time upgrade of intelligence video’s "brain". This research has a good popularization value in the application field of intelligent video technology.

Ključne riječi

convolutional neural network; deep learning technology; intelligent video; optical flow method

Hrčak ID:

199152

URI

https://hrcak.srce.hr/199152

Datum izdavanja:

21.4.2018.

Posjeta: 2.964 *