摘要
基于鞋样的视频追踪技术是公安机关刑事侦查的一种常用技战法,在公安实战中起到巨大的作用,然而该项技术大量依赖于人工筛选,工作量大且效率低,容易出现漏检的状况。鉴于此,提出一种基于SSD(Single Shot MultiBox Detector)模型的鞋子自动检测算法,实现对行人鞋子的自动检测与定位。首先对SSD模型的结构和先验框参数进行设计,使其符合鞋子检测的实际应用。然后采用调节网络参数的方法提高网络的检测性能和稳定性,完善适用于鞋子检测的网络模型和方法,最终得到准确且高效的单类别鞋子检测网络。最后在课题组前期构建的鞋样本数据库中进行性能评价。实验结果表明,所提算法的平均精度达到0.891。
Video tracking technology based on shoe patterns is commonly used by public security organizations in criminal investigations and is crucial in actual public security combat.However,this technology relies heavily on manual screening with a heavy workload and low efficiency,and it is prone to missed inspections.In view of this,an automatic shoe detection algorithm based on Single Shot MultiBox Detector(SSD)model is proposed herein to achieve automatic detection and positioning of pedestrian shoes.First,the structure of the SSD model and the parameters of the prior frame are designed to meet the practical application of shoe detection.Then,the method for adjusting network parameters is used to improve the detection performance and stability of the network and the network model and method suitable for shoe detection are improved.Thus,an accurate and efficient single-category shoe detection network is obtained.Finally,performance evaluation in the shoe sample database constructed by the research group in the early stage is conducted.Experimental results show that the average accuracy of the proposed algorithm is 0.891.
作者
耿鹏志
杨智雄
张家钧
唐云祁
Geng Pengzhi;Yang Zhixiong;Zhang Jiajun;Tang Yunqi(School of Criminal investigation,People's Public Security University of China,Beijing 100038,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2021年第6期176-183,共8页
Laser & Optoelectronics Progress
基金
国家重点研发计划(2017YFC0822003)
中国人民公安大学“公共安全行为科学研究与技术创新专项”项目。
关键词
图像处理
鞋子检测
卷积神经网络
SSD
视频侦查
image processing
shoes detection
convolutional neural network
Single Shot MultiBox Detector
video investigation
