Pedestrian retrieval enhancement method based on pedestrian detection, attribute learning and pedestrian recognition

Abstract

The invention discloses a pedestrian retrieval enhancement method based on pedestrian detection, attribute learning and pedestrian recognition. The method comprises the steps of pedestrian detection,pedestrian re-identification, pedestrian attribute prediction, pedestrian detection loss function, attribute classification loss function and identity classification loss function are used, attributeand identity labels are used to identify the position of a pedestrian in an image through a training framework, and a final loss function is obtained. According to the method, a multi-task deep learning framework is developed to solve the pedestrian retrieval problem, pedestrian detection, pedestrian re-identification and pedestrian attribute prediction are comprehensively considered in the framework in a single convolutional neural network, and the retrieval precision is improved.

Description

technical field [0001] The invention relates to a pedestrian retrieval enhancement method based on pedestrian detection, attribute learning and pedestrian identification. Background technique [0002] Pedestrian retrieval has received increasing attention due to its important role in video surveillance, which aims to retrieve persons of interest in multiple non-overlapping camera views. Given a picture, the task is to rank all pedestrian images candidates by calculating the similarity / distance between the picture and candidate images, and return the most relevant image as the retrieval result. It mainly consists of two parts: feature extraction and metric learning. The first part focuses on designing more robust features. The second part is to learn an appropriate distance / similarity function, using the features extracted from the image to better describe the similarity relationship between similar / different samples. In the early work, most of the work only used one featu...

AI Technical Summary

Problems solved by technology

On the downside, the semantic bias between the pose dataset and the pedestrian retrieval dataset can affect the accuracy of pedestrian retrieval.

[0003] On the other hand, due to the different information carried by different features, the use of a single feature cannot achieve better results.

Images