3D Vehicle Tracking Simulation Dataset

为了收集 3D 车辆跟踪模拟数据集，使用驾驶模拟来获得准确的 3D 边界框注释，而无需人工成本。数据收集和注释管道扩展了 VIPER 和 FSV 等以前的工作，特别是在跨帧链接身份方面。该模拟基于 Grand Theft Auto V，这是一款现代游戏，可在逼真的 3D 世界中模拟功能正常的城市及其周围环境。请注意，管道是实时的，提供了大规模数据收集的潜力，而 VIPER 需要昂贵的离线处理。To collect a simulated 3D vehicle tracking dataset, driving simulation was utilized to acquire accurate 3D bounding box annotations without manual labeling expenses. The data collection and annotation pipeline extends prior research including VIPER and FSV, with specific enhancements in identity association across consecutive frames. This simulation is based on Grand Theft Auto V (GTA V), a modern video game that faithfully recreates fully functional cities and their surrounding environments within a realistic 3D world. Notably, the proposed pipeline supports real-time operation, unlocking the potential for large-scale data collection, whereas VIPER requires costly offline processing.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于3D车辆跟踪的模拟数据集，通过驾驶模拟自动生成精确的3D边界框注释，无需人工成本。它基于《Grand Theft Auto V》游戏构建逼真的城市环境，由苏黎世联邦理工学院等机构于2021年发布，相关技术发表于2022年的IEEE期刊论文。