Action Detection Based On Faster RCNN

This paper studies action detection,designed to design an action detection method with high detection accuracy.In recent years,Convolutional Neural Networks(CNNs)have been developed in the fields of image recognition and video recognition.Our action detection method is also based on Convolutional Neural Network.The action detection method is divided into three stages.The first stage uses the Faster RCNN network to obtain multiple regions of interest at the frame level,and uses the network to extract the appearance features and motion features of each region of interest respectively.The second stage uses the Support Vector Machine(SVM)to classify the regions of interest at the frame level.The third stage connects the regions of interest with the classification of the high score and forms many action tubes,and then views the highest score of action tube as the final action detection result.In order to verify the effectiveness and higher detection accuracy of the method,we experimented on JHMDB and UCF Sports data sets,and compared with the current best action detection method constructed by Convolutional Neural Network.When the overlap threshold is 0.5,the detection accuracy of the action detection method is 60.86% and 80.12% respectively with 7.5 and 4.3 percentage points increase in the two data sets,and the detection accuracy is improved significantly.In this paper,we also show that the detection accuracy is 56.12% and 80.12% in the two data sets when the overlap threshold is 0.6,which is still higher than the current best method for an overlap threshold of 0.5;in addition,the accuracy of this method for an overlap threshold of 0.6 is still 80.12% on UCF Sports,which indicates that this method can achieve higher detection accuracy under the requirement of more accurate positioning.Higher detection accuracy means that the action recognition error is smaller and the action positioning is more accurate.The experimental results show that the method is effective and can achieve higher detection accuracy in the task of action detection in videos.The action detection method proposed in this paper has certain research and application value.