Action recognition method based on yolov3 and bag-of-words model

Abstract

The behavior recognition method based on YOLOv3 and bag-of-words model includes the following steps: read video frames, use YOLOv3 network for target detection, and return the position information of the target; intercept the target area and generate an action sequence; pre-predict the sequence frames in the action sequence process, and then extract multi-scale HOG features and SIFT features respectively; perform feature weighted fusion on the extracted multi-scale HOG features and SIFT features; use the K-means clustering algorithm to cluster the fusion features obtained after the weighted fusion in the previous step, Construct a visual dictionary; input the visual dictionary vector of the action sequence into the SVM multi-classifier model for training and recognition. This method uses the YOLOv3 network to detect the target and accurately intercepts the target area, and combines it with the bag-of-words model to reduce training parameters and background noise. The recognition rate on the KTH data set reaches 96.09%, which provides a new method for efficient and accurate recognition of video behaviors. Methods.

Description

technical field [0001] The invention relates to the field of action recognition, in particular to an action recognition method based on YOLOv3 and a bag of words model. Background technique [0002] In recent years, with the promotion of smart cities and the wider application of video surveillance in daily life, the subject of pedestrian object detection has received more and more attention in the field of computer vision, and it is also playing an increasingly important role in many scenarios. more important role. In the military field, it can be used for tasks such as criminal positioning and tracking, pedestrian analysis, etc. In the civilian field, it can be used for intelligent assisted driving, intelligent monitoring and other tasks. It can be said that the pedestrian target detection technology has provided great convenience to our life unconsciously. In practical engineering, people often need to detect and locate pedestrians on a large number of images or videos. ...

AI Technical Summary

Problems solved by technology

Although the classic machine learning algorithm has improved performance to a certain extent compared with the traditional method, it still has the problem of low detection accuracy for pedestrians with different postures and angles.

Images