| 期刊名称： |
Remote Sensing |
| 全部作者： |
Jiasong Zhu*,Ke Sun,Sen Jia,Weidong Lin,Xianxu Hou,Bozhi Liu,Guoping Qiu |
| 出版年份： |
2018 |
| 卷 号： |
10 |
| 期 号： |
6 |
| 页 码： |
887-908 |
| 查看全本： |
|
Vehiclebehaviorrecognitionisanattractiveresearchfieldwhichisusefulformanycomputervisionandintelligenttrafficanalysistasks.Thispaperpresentsanall-in-onebehaviorrecognitionframeworkformovingvehiclesbasedonthelatestdeeplearningtechniques.Unliketraditionaltrafficanalysismethodswhichrelyonlow-resolutionvideoscapturedbyroadcameras,wecapture4K(3840×2178)trafficvideosatabusyroadintersectionofamodernmegacitybyflyingaunmannedaerialvehicle(UAV)duringtherushhours.Wethenmanuallyannotatelocationsandtypesofroadvehicles.Theproposedmethodconsistsofthefollowingthreesteps:(1)vehicledetectionandtyperecognitionbasedondeepneuralnetworks;(2)vehicletrackingbydataassociationandvehicletrajectorymodeling;(3)vehiclebehaviorrecognitionbynearestneighborsearchandbybidirectionallongshort-termmemorynetwork,respectively.Thispaperalsopresentsexperimentalresultsoftheproposedframeworkincomparisonwithstate-of-the-artapproachesonthe4Ktestingtrafficvideo,whichdemonstratedtheeffectivenessandsuperiorityoftheproposedmethod.
