Abnormal Action Recognition Method Of Engine Driver Based On Deep Learning

With the rapid development of railway system,the running speed of locomotive has been improved several times,and the locomotive drivers put forward higher requirements.In addition,object detection and action recognition based on deep learning is the most popular research direction.This paper mainly studies the abnormal action of locomotive drivers based on deep learning.The purpose is to detect,identify and forewarn the abnormal actions of drivers that may cause accidents by using deep learning method.The main work of this paper is as follows.The image preprocessing algorithms are analyzed,including image enhancement,denoising and segmentation.Firstly,histogram equalization and fuzzy enhancement are mainly analyzed in image enhancement technology,and mean filter,wavelet filter and adaptive median filter are mainly analyzed in image denoising technology.Finally,the image segmentation technology is studied.In the image segmentation technology,the image segmentation technology based on threshold and clustering is analyzed.The action recognition algorithm based on traditional methods is analyzed.The IDT algorithm of traditional action recognition method is the best method outside the field of deep learning,so it mainly analyzes the three steps of dense sampling,track tracking and feature extraction.In addition,the correlation analysis and research of SVM classifier are carried out,and the experiment of correlation algorithm is carried out.This paper focuses on the detection and recognition of driver's abnormal action.First,the improved R-CNN model,Faster R-CNN,is analyzed and studied.Secondly,aiming at the problem of slow detection speed of R-CNN's series models,we focus on the YOLOv3 algorithm and the Retina Net algorithm,and put forward the improvement methods for the problems that affect the detection speed or detection accuracy.In addition,we also make the data set and define the abnormal action.We use the improved method of YOLOv3 and the improved method of Retina Net to train the model,and compare the two methods with Faster R-CNN,yolov3 and retinanet algorithm,and compare their advantages and disadvantages.We can choose according to the use environment.Finally,based on the existing training model,a detection and recognition system for the abnormal driving action of railway drivers is designed and developed,which realizes the collection and recognition analysis of abnormal action.