Person Re –Identification Based on DDN Foreground Extraction and Mapping Model Learning

With the wide application of video surveillance equipment, personre-identification has become a critical task in intelligent video surveillance and has a wide application prospect. This paper presents a person re-identification algorithm based on deep decompositional network foreground extraction and mapping model learning. Firstly, the DDN model is used to segment the pedestrian images,then the color histogramof the foreground images and the Gabor texture feature of the whole original images are extracted. The cross-view mapping model between the cameras will be studied by using the extracted pedestrian features. Finally, the person features of the search set and the candidate set are transformed into a same space where the feature distribution is more consistent, after that the distance metric learning and sorting are performed. Experiments show that the model can extract more robust person features, overcoming the back ground interference, and the person re-identification matching rate has been effectively improved.