带有语义信息的动态占用网格地图使用基于深度学习的BEVFusion方法与摄像头和LiDAR融合
Harin Jang1, Taehyun Kim1, Kyungjae Ahn1
1Graduate School of Automotive Engineering, Kookmin University, 77 Jeongneung-ro, Seongbuk-gu, Seoul 02707, Republic of Korea.
Sensors (Basel, Switzerland)
|May 11, 2024
概括
本研究引入了通过融合摄像头和LiDAR数据增强对象分类的动态占用网格图 (DOGM). 这提高了在复杂的城市环境中对自动驾驶汽车的看法.
科学领域:
- 机器人技术和自主系统
- 传感器融合式传感器
- 人工智能的人工智能
背景情况:
- 动态占用网格地图 (DOGM) 对于机器人和自动驾驶中的对象位置和速度的表示至关重要.
- 目前基于3D光检测和测距 (LiDAR) 的DOGM缺乏对象分类功能,限制了它们的应用范围.
- 将摄像头数据与LiDAR集成至关重要,以克服单传感器系统的局限性.
研究的目的:
- 开发一种基于深度学习的新型传感器融合技术,以增强DOGM的对象类信息.
- 提高自动驾驶汽车感知系统的可靠性和范围.
- 利用Dempster-Shafer证据理论将类信息和不确定性纳入DOGM.
主要方法:
- 实施了用于摄像机-LiDAR传感器融合的深度学习模型,作为DOGMs的输入.
- 根据Dempster-Shafer证据理论开发了更新规则,以整合对象类和不确定性.
- 调查了两个占用概率分配模型 (边缘与整个边界框) 用于速度估计准确性分析.
- 使用nuScenes数据集进行性能评估.
主要成果:
- 融合传感器方法成功地将对象类信息纳入DOGM中,扩大了它们的适用性.
- 未分类的LiDAR点为地图形成做出了贡献,提高了整体感知可靠性.
- 用不同的占用概率分配模型进行了速度估计准确性的分析.
- 开发的技术在nuScenes数据集上表现出有效的性能.
结论:
- 增强的DOGM与对象类信息为自动驾驶汽车提供了更丰富的感知数据.
- 这一进步对于在复杂的城市驾驶场景中实现更安全,更有效的导航至关重要.
- 深度学习,传感器数据和证据理论的融合为环境感知提供了强大的方法.
相关概念视频
01:15Depth Perception and Spatial Vision
631
Depth perception is the ability to perceive objects three-dimensionally. It relies on two types of cues: binocular and monocular. Binocular cues depend on the combination of images from both eyes and how the eyes work together. Since the eyes are in slightly different positions, each eye captures a slightly different image. This disparity between images, known as binocular disparity, helps the brain interpret depth. When the brain compares these images, it determines the distance to an object.
631
01:29Levels of Use of a GIS
49
Geographic Information Systems (GIS) operate across three levels of application, each representing an increasing degree of complexity: data management, analysis, and prediction. These levels reflect the expanding functionality and versatility of GIS technology in handling spatial data for diverse purposes.Data ManagementAt its foundational level, GIS serves as a tool for data management, enabling the input, storage, retrieval, and organization of spatial data. This level is often employed in...
49
01:23GIS Software, Hardware, and Sources of GIS Data
57
A Geographic Information System (GIS) combines specialized software and hardware to effectively manage, analyze, and present spatial and related data. GIS software includes critical functionalities such as a user interface for easy navigation, database management tools for handling spatial and attribute data, and data retrieval features for efficient access. Analytical tools transform raw data into insights, while display functions produce maps and reports in various formats for effective...
57
01:28Field Application of Global Positioning System
43
The Global Positioning System (GPS) has become an indispensable tool in fieldwork, offering unparalleled precision and efficiency for surveying, navigation, and infrastructure development. By harnessing signals from a constellation of satellites, GPS receivers determine the location of objects with remarkable speed and accuracy, often completing calculations within a second.Advantages of Modern GPS TechnologyContemporary GPS receivers are designed to meet the practical demands of field...
43