项目名称: 基于深度学习的交通环境理解与目标检测方法研究
项目编号: No.91320101
项目类型: 重大研究计划
立项/批准年度: 2014
项目学科: 计算机科学学科
项目作者: 乔宇
作者单位: 中国科学院深圳先进技术研究院
项目金额: 100万元
中文摘要: 本项目面向无人驾驶的行驶需求,借鉴生物视觉机理,特别是人视觉皮层信息处理的多层、逐层抽象、上下文融合等特性,利用近年来发展起来的深度学习方法作为主要工具,并以项目组近年在计算机视觉和智能交通领域取得的各项成果为基础,研究交通环境的解析和理解方法,研究车道的检测、路牌的检测与识别、交通场景的分类等方法,研究基于深度学习的行人和车辆检测方法。我们将利用真实数据和国际公开数据库,对所开发的方法进行测试。在此基础上,进行集成验证。
中文关键词: 深度学习;场景理解;目标检测;路牌识别;
英文摘要: Aiming at the requirements of unmanned driving and inspired the biology vision mechanism such as hierarchy, abstraction through multi-layer processing and contextual integration, this project makes use of deep learning as a main tool and study the following methods based on our recent research results in computer vision and intelligent transportation fields: traffic scene parsing and understanding, lane detection, traffic sign detection and recognition, traffic scene classification, and human and vehicle detection. We will examine and evaluate the developed methods on real data and international public datasets. We will also conduct system integration and verification.
英文关键词: Deep learning;Traffic scene understanding;Object detection;Traffic sign recognition;