Optimizing Moving Object Trajectories from Roadside Lidar Data by Joint Detection and Tracking

High-resolution traffic data, comprising trajectories of individual road users, are great importance to the development Intelligent Transportation Systems (ITS), in which they can be used for microsimulations and applications such as connected vehicles. Roadside laser scanning systems increasingly being tracking on-road objects, tracking-by-detection is widely acknowledged method; however, this method sensitive misdetections, resulting shortened discontinuous object trajectories. To address this, a Joint Detection And Tracking (JDAT) scheme, runs detection parallel, proposed mitigate miss-detections at vehicle stage. Road users first separated by moving point semantic segmentation then instance clustering. Afterwards, two procedures, tracking, conducted parallel. In detection, PointVoxel-RCNN (PV-RCNN) employed detect vehicles pedestrians from extracted points. tracker utilizing Unscented Kalman Filter (UKF) Probabilistic Data Association (JPDAF) obtain all objects. The identities determined results using only certain number representatives each trajectory. developed scheme has been validated three urban study sites different lidar sensors. Compared with method, average range increased >20%. approach also successfully maintain continuity bridging gaps caused miss-detections.