基于ReInspect算法的多目标追踪  

Multi-target Tracking Algorithm Based on ReInspect Algorithm

作　　者：王文远 金晅宏[1] 宋文净 王轶炜 WANG Wen-yuan;JIN Xuan-hong;SONG Wen-jing;WANG Yi-wei(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《计量学报》2022年第4期470-474,共5页Acta Metrologica Sinica

摘　　要：为了提高复杂场景下多目标检测的准确性,提出了一种基于ReInspect算法的对于多个运行目标的检测方法。该算法基于OverFeat算法和Faster R-CNN算法的思想,加入LSTM(long short-term memory)循环网络结构用于记录多个目标的特征序列;通过调整LSTM网络特征标签信息,预处理损失函数,并在追踪后采用置信度分段的方式对检测结果进行匹配,解决对同一目标的重复检测以及目标遮挡问题。实验结果表明,该算法对传统的重叠、遮挡等问题有较好的抗干扰能力,在不同场景下识别准确率均高于90%。In order to improve the accuracy of multi-target detection in complex scenes,a detection method for multiple moving objects based on ReInspect algorithm is proposed.The algorithm is based on the OverFeat algorithm and Faster R-CNN algorithm,adding LSTM(long short-term memory)recurrent network structure to record the feature sequence of multiple targets.By adjusting the LSTM network feature label information to preprocess the loss function,and the confidence segmentation method is used after tracking to match the detection results to solve the problem of repeated detection of the same target and target occlusion.The experimental results show that the algorithm has good anti-interference ability against traditional overlapping and occlusion problems,and the recognition accuracy is higher than 90%in different scenarios.

关 键 词：计量学 多目标追踪 ReInspect算法 深度学习 目标检测 

分 类 号：TB96[机械工程—光学工程]