视频监控领域深度特征编码的行人检测算法

(1.阿坝师范学院 数学与计算机科学学院,四川 汶川 623002; 2.西华师范大学 数学与信息学院,四川 南充 637002)

模式识别; 深度编码; 行人检测; SVM算法; 自编码网络; 聚合通道特征

Deep feature coding for pedestrian detection in video surveillance
LUO Nan-chao1,ZHENG Bo-chuan2

(1.School of Mathematics and Computer Science,Aba Teachers University,Wenchuan 623002,China; 2.School of Mathematics and Information,China West Normal University,Nanchong 637002,China)

pattern recognition; deep coding; pedestrian detection; SVM Model; auto-encoder network; aggregation channel feature

DOI: 10.13800/j.cnki.xakjdxxb.2019.0419

备注

由于高清视频监控领域现有行人检测算法在复杂背景下检测准确率不高且检测实时性不强,提出了一种新颖的深度特征行人检测算法,该算法利用聚合通道特征模型对监控高清图像进行预处理,筛选出具有显著特性的疑似目标,大大降低目标检测的数量; 然后对获取的疑似目标区域进行尺度校正与特征提取,并输入到深度模型中进行深度特征编码,提高特征的表征能力; 最后输入到LSSVM分类模型,得到最终的行人检测结果。仿真实验结果显示所提行人检测算法在保证检测准确率的同时,具有较高的检测效率。

To solve the problem of poor real-time detection and low precision in video surveillance,a novel deep feature-based pedestrian detection algorithm is proposed.The algorithm firstly uses the aggregation channel feature model to process the surveillance images,and selects the suspected target region with salient characteristics.Then,the scaled correction and feature extraction are performed on the obtained suspected target region.The corresponding low-level features are obtained and input into the deep auto-encoder network for deep feature coding so as to enhance the representation ability.Finally,the coding feature is input into the least squares SVM classification model to obtain the final detection results.A large number of qualitative and quantitative experimental results show that the proposed detection algorithm guarantees the accuracy of pedestrian detection with higher efficiency.