基于遮挡目标去除的公交车拥挤度分类算法  

Bus crowdedness classification algorithm based occluded object removal

作　　者：孟令灿 聂秀山 张雪 MENG Lingcan;NIE Xiushan;ZHANG Xue(School of Computer Science and Technology,Shandong Jianzhu University,Jinan 250000,Shandong,China)

机构地区：[1]山东建筑大学计算机科学与技术学院,山东济南250000

出　　处：《山东大学学报（工学版）》2022年第4期83-88,共6页Journal of Shandong University（Engineering Science）

摘　　要：由于公交车中场景复杂、干扰因素繁多容易出现遮挡乘客问题,现有深度学习和目标检测方法在对公交车内的拥挤程度分类时精度低、效果差,往往达不到令人满意的效果。针对这一问题,提出一种基于遮挡目标去除的公交车拥挤度分类算法,对公交拥挤进行分类和分析。该方法有遮挡物检测、图像去遮挡和拥挤度分类模块三部分组成。基于目标检测算法检测出遮挡物,通过图像修复算法对乘客图像进行修复,利用拥挤度分类算法分析拥挤度。本研究从真实的公交车中采集数据生成数据集,并进行标注。试验结果表明,基于遮挡目标去除的分类算法的准确率达到了67.12%,与现有的方法对比具有最高的预测精度。Due to the complex scenes and numerous interference factors in buses,the problem of shielding passengers was easy to occur,leading to the low accuracy and poor effect of existing deep learning methods when classified the crowding degree in buses,which often failed to achieve satisfactory results.To address this issue,a bus crowdedness classification algorithm based occluded object removal was proposed to classify and analyze bus crowding.This proposed method consisted of three parts,containing the occlusion object detection,image de-occlusion and crowdedness classification modules.The occlusions would be identified by occlusions object detection algorithm,and then inpaint the image of passengers via the image inpainting algorithm.The degree of crowding was detected using the crowd detection algorithm.This study adopted a dataset collected from the real bus with annotation.The experimental results showed that the accuracy of the classification algorithm based on occlusion object removal reached 67.12%,which had the highest prediction accuracy compared with the existing methods.

关 键 词：拥挤分类 目标检测 拥挤图像 图像修补 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]