Research And Application Of Video Content Understanding

With the wide application of the Internet and the development of modern information processing technology,video data shows an explosive growth trend.Because of the complex structure,rich content and unstructured characteristics of video data,people have limited ability to process massive video data,and the potential information has not been fully mined.Therefore,more intelligent technology is needed to process the video data.Video content understanding is not only the main means of intelligent video processing,but also a research hotspot and difficulty in the field of computer vision.The involved subjects are pattern recognition,image processing,computer vision,artificial intelligence.It has great significance,wide application prospects and potential economic value in military,civil and medical fields.This paper mainly realizes video content understandin g based on video key frame extraction and object detection,and elaborates video content understanding technology and application through the practical application of video content understanding in automatic driving.The work of this paper consists of the following five parts:(1)In view of the unstructured and difficult characteristics of video data,HSV histogram method is used to transform abstract and complex high-dimensional data into quantifiable low-dimensional data,thus reducing the amount of data.(2)Considering the high similarity of adjacent frames in video data,key frame extraction is transformed into clustering problem.K-Means,AGNES and density peak clustering algorithms are designed and implemented to extract video key frames,and their clustering effects are analyzed.Then,the results of key frame extraction algorithms in compressed domain and uncompressed domain are compared,and a video key frame extraction algorithm with better comprehensive performance is obtained.(3)In order to ensure the quality of clustering,Silhouette Coefficient is used to calculate the optimal cluster number to determine the initial cluster center and the number of clusters.(4)In order to improve the detection accuracy of object detection model,the data set is pruned to make the model better adapt to specific application scenarios.Experiments show that the improved model has improved the recognition accuracy.(5)Combined with the key frame extraction and object detection algorithm,a complete video understanding experiment process is constructed.The experimental data are analyzed under the background of the autopilot system,and the practical application of video content understanding in automatic driving is presented.