Recently, Dahua Technology’s Person Re-identification based on deep learning algorithm once again created a new historical record on the three major international authoritative public datasets: Market1501, DukeMTMC and CUHK03.
This is the third time that Dahua AI Technology has made breakthrough progress on these data sets since 2018 and 2019. With a new high accuracy rate, Dahua Technology continues to maintain the lead in the person re-identification field.
Comparison results on public datasets at two key Indicators: Rank1 accuracy and mean average precision
About Person Re-identification
Person Re-identification aims to match the images of the same person across multiple camera views. It faces challenges such as occlusion, illuminations and articulated deformations of the human pose. With video surveillance applications such as cross-camera person tracking and person search, Person Re-identification plays an important role in public security and criminal investigation.
Based on the researches and long-term accumulation of experience on person re-identification, Dahua Technology has developed a novel frame of training and testing, mainly solving domain gap, views changing, occlusions and deformations of the human pose. By improving key technical strategies, the problems of occlusion, misalignment and partial person retrieval in complex scenarios were solved, and the effect was remarkable.
At present, this technology has been successfully applied to Dahua video structured cameras, NVR storage, intelligent video metadata servers and other products. Through comprehensive upgrading of product capabilities, it has enabled smart applications in public security, transportation, finance, business and other industries, significantly improving work efficiency and video applications, as well as constantly creating new value for customers. In the Hangzhou Safe City Project, the Dahua AI Person Re-identification Technology plays an important role in helping the police to quickly recover lost elderly people and children.
