Vehicle detection and tracking dataset in complex road surveillance scenarios
Aiming at the problems of single acquisition scenario,long-tailed distribution of datasets,and simple image environment in most existing vehicle detection and tracking datasets,a vehicle dataset VeDT-MSS(vehicle detection and tracking for multiple surveillance scenarios)is proposed in this paper for the detection and tracking research of four vehicle categories(car,truck,bus,and motorcycle)in urban and rural surveillance scenarios.The dataset has four salient features:diverse traffic scenarios,large intra-class diversity of trucks,an elevated proportion of motorcycle annotation instances,and high background complexity.To validate the effectiveness of this dataset,a large number of baseline experiments have been performed on object detection and multi-object tracking tasks.Experimental results demonstrate the utility of the VeDT-MSS dataset in evaluating the robustness and generalization of existing algorithms.The proposed dataset has considerable potential to facilitate vehicle detection and tracking research and to provide the computer vision community with a new selection of data for evaluating algorithm performance.
vehicle detection and trackingdatasetsurveillance scenarioVeDT-MSSdeep learningobject detectionmulti-object trackingrural road
万方数据
维普
