中文摘要：

为满足先进驾驶辅助系统的高准确性行人检测要求,提出一种模拟人类注意力机制的视觉显著性行人检测方法。基于仅含行人信息的标记样本库,建立了条件随机场（CRF）模型,以实现不同显著性计算方法的最优融合。实际检测中,首先采用SLIC算法进行图像超像素形式的几何信息划分,进而对可能存在行人的区域进行初筛选,随后在可能的行人区域内,采用CRF模型计算显著性,并将具有较高显著性的区域确定为行人区域。实验结果表明,该方法具有较好的判别性能并达到满意的检测率,同时,采用的行人区域筛选方法在一定程度上缩短了算法的检测时间,基本满足了车载平台的实时性要求。

英文摘要：

For meeting the requirements of high accuracy of pedestrian detection in advanced driver assistance systems,a visual saliency based pedestrian detection method is proposed to simulate human attention mechanism.Based on the labeled sample bank containing only pedestrian information,a conditional random field（CRF）model is set up to achieve the optimal fusion of different saliency calculation methods.In practical detection,the SLIC algorithm is used firstly to divide the image geometric information into super pixels so that the regions probably having pedestrian can be preliminarily selected.Then the CRF model is used to calculate the saliency of probable pedestrian regions and the regions with high saliency are determined to be pedestrian regions.Experiment results show that the method proposed has good discrimination performance with satisfactory detection rate.In addition,the pedestrian region searching method used reduces the detection time of algorithm to a certain extent,basically meeting the real-time requirements of onboard platform.