Bi-directional Interaction Network for Person Search Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing person search methods suffer from low accuracy due to the influence of scene information outside the person's boundary on feature extraction, which hinders effective identity identification.
Innovation Solution
A bi-directional interaction network (BINet)-based person search method is introduced, utilizing a residual network with a new classification layer and an instance-aware branch to focus on person-specific information, eliminating irrelevant scene information by bi-directional information interaction during training, and using a pre-trained person search model to enhance feature discrimination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a deep convolutional network is used to extract person features from scene images, then the receptive field increases and more scene information is captured, but the discrimination of identity features becomes insufficient due to excessive scene information interference
Solution Approach 1:
The network is divided into two separate branches: a detection branch that processes full scene images for person localization, and an instance-aware branch that processes cropped person regions for identity recognition. This segmentation allows each branch to specialize in its specific task without interference from irrelevant scene information.
Solution Approach 2:
The instance-aware branch extracts and processes only the person region of interest from the scene image, separating the identity feature extraction task from the surrounding scene information. This extraction enables focused processing of person-specific features without scene interference.
2Productivity
If a multi-task model is used to complete person detection and identity identification simultaneously, then the system integrates both functions in one stage, but the scene information outside person boundary negatively affects identity identification accuracy
Solution Approach 1:
The unified model is segmented into two parallel branches with distinct functions: the detection branch handles scene-level person localization, while the instance-aware branch handles person-level identity recognition. This architectural segmentation resolves the conflict between efficient one-stage processing and accurate identity identification.
Solution Approach 2:
The instance-aware branch acts as an intermediary that receives person region proposals from the detection branch and processes them separately for identity recognition. This intermediary structure allows the system to maintain efficient one-stage processing while isolating identity features from scene information interference.
Data Source
AI summary
A bi-directional interaction network (BINet)-based person search method, system, and apparatus are provided. The method includes: obtaining, as an input image, a tth frame of image in an input video; and normalizing the input image, and obtaining a search result of a to-be-searched target person by using a pre-trained person search model, where the person search model is constructed based on a residual network, and a new classification layer is added to a classification and regression layer of the residual network to obtain an identity classification probability of the target person. The method improves the accuracy of the person search.