{"title":"无人机视频分析的模型预测压缩","authors":"Aakanksha Chowdhery, M. Chiang","doi":"10.1109/SECONW.2018.8396351","DOIUrl":null,"url":null,"abstract":"Drones will be increasingly deployed in surveillance scenarios, disaster zones, and remote areas. The videos collected from drone cameras provide site surveys, summaries, detect and track multiple targets. Today such videos are processed offline after the drone flight. Real- time processing provides several opportunities but has two key challenges - network bandwidth on the drone-to-server link is constrained and the computational capability of the drone processor is limited in terms of applying machine vision in real time. We propose a model predictive compression algorithm that uses predicted drone trajectory to select and transmit the most important image frames to the ground station to maximize the application utility while minimizing the network bandwidth use. The proposed compression scheme works in real-time on the drone processor because it estimates background motion without computing image features. To correct the model inaccuracies, the drone receives feedback from the ground station that can compute image features in real time. Evaluation results suggest that the proposed compression approach reduces network bandwidth overheads by 50-72 % while ensuring high-quality mosaics in the drone mosaicing application.","PeriodicalId":346249,"journal":{"name":"2018 IEEE International Conference on Sensing, Communication and Networking (SECON Workshops)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2018-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"10","resultStr":"{\"title\":\"Model Predictive Compression for Drone Video Analytics\",\"authors\":\"Aakanksha Chowdhery, M. Chiang\",\"doi\":\"10.1109/SECONW.2018.8396351\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Drones will be increasingly deployed in surveillance scenarios, disaster zones, and remote areas. The videos collected from drone cameras provide site surveys, summaries, detect and track multiple targets. Today such videos are processed offline after the drone flight. Real- time processing provides several opportunities but has two key challenges - network bandwidth on the drone-to-server link is constrained and the computational capability of the drone processor is limited in terms of applying machine vision in real time. We propose a model predictive compression algorithm that uses predicted drone trajectory to select and transmit the most important image frames to the ground station to maximize the application utility while minimizing the network bandwidth use. The proposed compression scheme works in real-time on the drone processor because it estimates background motion without computing image features. To correct the model inaccuracies, the drone receives feedback from the ground station that can compute image features in real time. Evaluation results suggest that the proposed compression approach reduces network bandwidth overheads by 50-72 % while ensuring high-quality mosaics in the drone mosaicing application.\",\"PeriodicalId\":346249,\"journal\":{\"name\":\"2018 IEEE International Conference on Sensing, Communication and Networking (SECON Workshops)\",\"volume\":\"29 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2018-06-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"10\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2018 IEEE International Conference on Sensing, Communication and Networking (SECON Workshops)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/SECONW.2018.8396351\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2018 IEEE International Conference on Sensing, Communication and Networking (SECON Workshops)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/SECONW.2018.8396351","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Model Predictive Compression for Drone Video Analytics
Drones will be increasingly deployed in surveillance scenarios, disaster zones, and remote areas. The videos collected from drone cameras provide site surveys, summaries, detect and track multiple targets. Today such videos are processed offline after the drone flight. Real- time processing provides several opportunities but has two key challenges - network bandwidth on the drone-to-server link is constrained and the computational capability of the drone processor is limited in terms of applying machine vision in real time. We propose a model predictive compression algorithm that uses predicted drone trajectory to select and transmit the most important image frames to the ground station to maximize the application utility while minimizing the network bandwidth use. The proposed compression scheme works in real-time on the drone processor because it estimates background motion without computing image features. To correct the model inaccuracies, the drone receives feedback from the ground station that can compute image features in real time. Evaluation results suggest that the proposed compression approach reduces network bandwidth overheads by 50-72 % while ensuring high-quality mosaics in the drone mosaicing application.