Enhanced pedestrian attribute recognition method based on group optimization reward

Abstract

The invention provides an enhanced pedestrian attribute recognition method based on group optimization rewards, and belongs to the field of machine learning. The method comprises the following steps: firstly, constructing a proper Markov decision process, extracting features of pedestrian images by using a CNN network, extracting attribute features by using a text vectorization tool, and taking a tuple formed by the features as a state of the Markov decision process; taking a set composed of 0 and 1 as an action space; and designing a state transition process and a reward function. A reinforcement learning algorithm is used for training, a state is input into a DQN network, and an attribute grouping strategy and a group optimization reward function are used for optimizing training of the network, so that an intelligent agent with a better strategy is obtained, and a pedestrian attribute recognition result is improved.

Description

technical field [0001] The invention relates to a pedestrian attribute identification method, in particular to an enhanced pedestrian attribute identification method based on group optimization reward. Background technique [0002] In recent years, video surveillance has been gradually improved in public places. Video surveillance equipment on the order of tens of millions collects video and image information of road traffic, pedestrians, etc. at all times. Monitoring the massive information obtained to improve the security level has gradually become an important research topic. Pedestrian attribute recognition is a very important task in surveillance video image analysis. Its purpose is to predict the attributes of pedestrian images obtained in surveillance videos to obtain global attributes such as age and gender and local attributes such as hats and coats. On the one hand, the pedestrian attribute recognition task can identify potentially dangerous people or objects from...

Images