High-resolution traffic data, comprising trajectories of individual road users, are of great importance to the development of Intelligent Transportation Systems (ITS), in which they can be used for traffic microsimulations and applications such as connected vehicles. Roadside laser scanning systems are increasingly being used for tracking on-road objects, for which tracking-by-detection is the widely acknowledged method; however, this method is sensitive to misdetections, resulting in shortened and discontinuous object trajectories. To address this, a Joint Detection And Tracking (JDAT) scheme, which runs detection and tracking in parallel, is proposed to mitigate miss-detections at the vehicle detection stage. Road users are first separated by moving point semantic segmentation and then instance clustering. Afterwards, two procedures, object detection and object tracking, are conducted in parallel. In object detection, PointVoxel-RCNN (PV-RCNN) is employed to detect vehicles and pedestrians from the extracted moving points. In object tracking, a tracker utilizing the Unscented Kalman Filter (UKF) and Joint Probabilistic Data Association Filter (JPDAF) is used to obtain the trajectories of all moving objects. The identities of the trajectories are determined from the results of object detection by using only a certain number of representatives for each trajectory. The developed scheme has been validated at three urban study sites using two different lidar sensors. Compared with a tracking-by-detection method, the average range of object trajectories has been increased by >20%. The approach can also successfully maintain continuity of the trajectories by bridging gaps caused by miss-detections.
1 June 2017 in IEEE Geoscience and Remote Sensing Letters
With the continuous improvement of image resolution, details on aerial images provide abundant available information for vehicle detection. Nevertheless, traditional works mainly exploited the overall information of the vehicles ignoring the local details, such as front and rear windshields, and thus, there were usually more than 15% false alarms in the final vehicle detection results. In this letter, we propose a vehicle detection method making full use of high level details on aerial images. In the training stage, we choose front windshield samples to train a part detector and whole vehicle samples to train a root detector. In the matching stage, we first use the root detector to define an entire vehicle obtaining the root response, then the part detector is scanned in the root bounding box to decide a front windshield and get the part response. Afterward, the part response is transformed by setting weight w based on the part position offset. More importantly, contextual information is appropriately used in the process of determining the part position offset. Final detection score is the combination of root response and the transformed part response. We have demonstrated that the proposed method has achieved better performance with more than 6.43% increase of correct detection rate and more than 5.63% decrease of false detection rate compared with the state-of-the-art approaches.