Deep learning model training data set construction method and system for violent behavior detection

Abstract

An embodiment of the invention discloses a deep learning model training data set construction method and a deep learning model training data set construction system for violent behavior detection. Thedeep learning model training data set construction method comprises the steps of: 1, storing and cleaning videos in a public data set; 2, storing and cleaning videos in a website; 3, recording a video containing violent behaviors; 4, recording a video based on an existing model; 5, setting an input frame number; 6, if a to-be-processed video exists, taking out the to-be-processed video; step 7, acquiring the total frame number of the video; step 8, calculating a frame taking interval; 9, calculating serial numbers of the taken frames; 10, naming pictures corresponding to the frame serial numbers according to specific parameters, and storing the pictures in a specified file system; and step 11, recording the to-be-processed video as 'processed'. According to the deep learning model training data set construction method, the input sample number and sample quality of a violent behavior detection model based on deep learning are improved, the training time is shortened, the resource overhead is reduced, the model performance is improved, and the violent behavior detection effect is further improved.

Description

technical field [0001] The present invention relates to the technical field of computer software applications, in particular to a method and system for constructing a deep learning model training data set for violent behavior detection. Background technique [0002] Artificial intelligence (AI), especially AI methods and technologies based on deep learning, has shown excellent performance in solving some problems in the fields of computer vision, speech recognition, natural language processing, etc., and has received extensive attention in recent years. In the field of machine vision, the designed deep learning model often has a huge number of parameters, and its training depends on a large number of data samples. For example, Facebook's facial recognition system DeepFace uses 4 million facial images from more than 4,000 identities. In 2017, Jeff Leek pointed out that if the amount of data you have is not large enough, you should not use deep learning methods. [0003] Vio...

AI Technical Summary

Problems solved by technology

If all the video frames of each video are input into the RNN model, a total of 750,000 video frames need to be input, which will cause high time and resource overhead

Images