基于生物视觉启发特征和遮挡模型的复杂道路环境目标检测方法研究

项目名称： 基于生物视觉启发特征和遮挡模型的复杂道路环境目标检测方法研究

项目编号： No.61503349

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 刘玮

作者单位： 中国地质大学（武汉）

项目金额： 22万元

中文摘要： 针对复杂道路环境感知中目标检测率低、误检率高以及遮挡目标难以准确检测的问题，本项目拟从目标的局部特征描述和遮挡状态描述两个方面展开研究，提高道路目标感知的鲁棒性。首先，针对复杂环境道路目标类间低分辨性和类内多样性，从生物视觉启发的角度出发，提出一种新的二层结构目标局部特征提取模型，并在此基础上构造局部描述子。该方法通过深入模拟视觉神经中简单细胞和复杂细胞的响应特性，得到目标更具鲁棒性的局部特征描述，从而提高目标检测率及降低误检率。其次，针对无先验知识情况下道路被遮挡目标难以检测的问题，根据道路目标各种遮挡状态之间的上下文关系，提出一种基于条件随机场结构的目标遮挡图模型，从而实现道路遮挡目标的准确检测。研究成果将为复杂道路环境感知中目标检测的难点问题提供新的有效解决思路和方法，具有重要理论意义和实际应用价值。

中文关键词： 目标检测；道路环境感知；生物启发特征；条件随机场；遮挡模型

英文摘要： This project will pursue research on local feature description and occlusion model of road object, in view of the problem of low detection rate of road object and error detection of occlusion object in urban environment perception. First, a novel biologically inspired local description will be proposed based on the visual information processing mechanism of ventral pathway in human brain. Features are extracted by a two-layer network, which works in a similar way as the simple cell and complex cell get responses. Local feature descriptors suppose to effectively represent local regions in a robust way to improve the detection rate of road object. Secondly, the project will model the occlusion state, which are defined on visibility of objects' parts, based on conditional random field. Parts contextual information will be used in occlusion model in order to detect occluded road object without prior knowledge. The objective of this project is to develop a new and effective way of solving difficult problems of urban road environment perception field. This project will have significant theoretical and practical results.

英文关键词： object detection;road environment perception;biologically inspired featres;conditional random field;occlusion model