Pedestrian attribute recognition method based on multi-temporal attention model

Abstract

The invention discloses a pedestrian attribute recognition method based on a multi-temporal attention model. The pedestrian attribute recognition method comprises the steps: obtaining image features and attribute features; constructing text supervision features, fusing two combination results of the image features and the attribute features, and cascading the attribute features to serve as attribute supervision; constructing a multi-temporal attention mechanism, namely constructing an alignment model of the attention mechanism by utilizing hidden layer vectors at two moments, and then jointlycarrying out weight optimization on image features; taking the text supervision feature and the context vector as additional input of a long-term and short-term memory model to obtain a hidden layer vector containing pedestrian attribute information; and obtaining a pedestrian attribute recognition probability; and optimizing the pedestrian attribute recognition probability. The pedestrian attribute recognition method can quickly and effectively recognize the attributes of different pedestrians in a real monitoring scene, plays an important role in promoting other deep learning fields, such aspedestrian retrieval and pedestrian re-recognition, and also plays a great positive role in building a safe city and perfecting a city monitoring system.

Description

technical field [0001] The invention relates to a pedestrian attribute recognition method. In particular, it involves a method for pedestrian attribute recognition based on a multi-temporal attention model. Background technique [0002] In modern cities, there are millions of surveillance cameras collecting video and picture information of pedestrians and traffic at all times. In order to protect people's lives and property and urban safety, it is necessary to analyze these massive data in real time. Early The monitoring system requires manual screening of data, which consumes a lot of manpower and material resources. With the development of machine learning, especially the rise of deep learning, it has become more convenient to process massive amounts of data. The task of pedestrian attribute recognition aims to predict the attributes of pedestrians, such as gender, age, clothing type, etc., when given an image containing pedestrians. It is important for processing massiv...

AI Technical Summary

Problems solved by technology

At present, there are special situations in the pedestrian attribute recognition task, that is, some attributes that account for a small proportion of the whole have relatively high recognition accuracy. On the contrary, some attributes that account for a large proportion of the whole have relatively low recognition accuracy.

Images