复杂道路监控场景下的车辆检测与跟踪数据集

Vehicle detection and tracking dataset in complex road surveillance scenarios

针对现有大部分车辆检测与跟踪数据集通常存在的采集场景单一、数据集长尾分布以及图像采集环境简单等问题,本文构建一个车辆数据集VeDT-MSS,用于城市以及乡村监控场景下4种车辆类别(小汽车、卡车、公交车和摩托车)的检测以及跟踪研究。该数据集具有交通场景多样化、卡车的类内多样性大、摩托车标注实例占比高以及背景复杂程度高4个显著特性。为了验证该数据集的有效性,在目标检测以及多目标跟踪任务上进行了大量的基线实验。实验结果表明,VeDT-MSS数据集在评估现有算法的鲁棒性和泛化性方面具有实用性。该数据集的提出对促进车辆检测与跟踪研究具有相当的潜力,并为计算机视觉社区评估算法性能提供一个新的数据选择。

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.