A dynamic target detection method and system

By constructing a topological complex and dynamic spatiotemporal graph network based on the theory of continuous cohomology, the problem of ignoring the topological structure features of event flow in existing technologies is solved, and high-precision and robust target detection in complex dynamic scenarios is achieved.

CN122435568APending Publication Date: 2026-07-21HUAIYIN INSTITUTE OF TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HUAIYIN INSTITUTE OF TECHNOLOGY
Filing Date
2026-04-21
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing GNN-based event camera target detection techniques neglect the deep topological features within the event flow when constructing the graph structure, failing to accurately capture topological evolution information, resulting in insufficient target detection accuracy and robustness in complex and dynamic scenes.

Method used

By constructing a topological complex based on continuous cohomology theory, extracting topological feature vectors, generating a dynamic spatiotemporal graph, and using a spatiotemporal graph attention network and a topological evolution prediction network to capture the dynamic evolution of the topological structure of the event flow, and combining density clustering algorithm to generate candidate anchor points for target detection.

Benefits of technology

It maintains high accuracy and robustness in complex dynamic scenarios, and can stably track and detect targets, especially when the target accelerates rapidly, turns sharply, or is occluded, thus improving detection accuracy and system stability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122435568A_ABST
    Figure CN122435568A_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of computer vision, and provides a dynamic target detection method and system, the method comprising: S1. Based on the event camera, event stream data is collected and denoising preprocessing is performed; S2. The preprocessed event stream data is mapped into a three-dimensional space-time point cloud, a topological complex is constructed, and a topological feature vector is extracted; S3. When the comprehensive connection edge weight between two event points is greater than a preset edge weight threshold, a dynamic space-time graph is generated; S4. The dynamic space-time graph and the original attribute feature vector and the topological feature vector associated with each event point are input into a pre-constructed space-time graph attention network to obtain high-order node features; S5. Motion trend features are obtained and are deeply fused with the high-order node features to generate candidate anchor points; S6. The detection result is obtained according to the candidate anchor points. Under complex dynamic changes, the application can still maintain stable, continuous and high-precision tracking and detection, and can be used to improve the driving safety of an autonomous vehicle in a complex environment.
Need to check novelty before this filing date? Find Prior Art