Visual Inertial Odometry Sensor Fusion Approach for Autonomous Localization

This paper describes a sensor fusion technique to localize autonomously unmanned vehicles. In particular, we performed a sensor fusion based on the extended Kalman filter between two commercial sensors. The adopted sensors are ZED2 and Intel T265, respectively; these platforms already perform visual-inertial odometry in their integrated system-on-chip. Since these 2 devices represent the top of the range on the market to make an autonomous localization, this study aims to analyze and inform about results that can be obtained by performing a sensor fusion between the two cameras. Several tests on a specific trajectory and environment demonstrated that a more robust autonomous localization than one of the single cameras can be obtained by properly tuning parameters and inputs of the Extended Kalman filter.

本文详细阐述了旨在实现自主无人车辆定位的一种传感器融合技术。具体而言，我们通过对两款商用传感器——ZED2与Intel T265——的扩展卡尔曼滤波进行融合。所选用的传感器已在其集成片上系统中实现了视觉惯性里程计。鉴于这两款设备在市场上代表着实现自主定位的顶尖水平，本研究旨在分析和报告通过在两个相机之间执行传感器融合所能获得的结果。在特定轨迹和环境下的多次测试表明，通过恰当地调整扩展卡尔曼滤波器的参数和输入，可以实现对单摄像头定位效果的显著增强，从而实现比单一摄像头更为稳健的自主定位。