Visual target tracking method, system and device based on multi-modal fusion and storage medium

By decoupling the visual target tracking task into label optical flow tracking and target tracking subtasks, and introducing high-precision visual label detection, a multi-hypothesis fusion strategy and a closed-loop feedback mechanism are adopted to solve the accuracy and stability problems of visual target tracking in complex environments, thus achieving high-precision and stable target tracking.

CN122416338APending Publication Date: 2026-07-17SOUTHEAST UNIV
0 Cites 1 Cited by

Patent Information

Application Number
CN202610570305.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-28
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing visual target tracking methods are not accurate enough or prone to drift when faced with target deformation, occlusion, and changes in lighting. Furthermore, the multi-sensor fusion strategy is simple and lacks an effective feedback mechanism, which leads to a decline in system performance in complex environments.

Method used

The visual target tracking task is decoupled into two parallel subtasks: tag optical flow tracking and target tracking. High-precision visual tag detection is introduced as an anchor point, and dynamic correction is performed through a fusion filter. A multi-hypothesis fusion strategy and a closed-loop feedback mechanism triggered by visual tag detection are adopted to achieve reliability assessment and intelligent fusion of the tracking source.

Benefits of technology

It improves the stability and adaptability of visual target tracking, ensuring high-precision target tracking in complex environments. It suppresses tracking drift through visual marker detection and provides intuitive feedback on position reliability.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a visual target tracking method, system and device based on multi-modal fusion, and a storage medium. The method comprises the following steps: initializing a video sequence, labeling a static or quasi-static reference label and a to-be-tracked target, extracting label sparse feature points, and establishing a conversion relation between pixels and real space coordinates; the main cycle tracks a reference label for each subsequent frame based on an optical flow method to obtain a stable position, parallelly obtains a target visual mark detection discrete position observation value and a continuous tracking observation value, and resets a tracker when the detection succeeds; a multi-hypothesis fusion strategy is adopted to update the state of a fusion filter, target position estimation is output, coordinate conversion is completed in combination with a stable position of a reference label, real space coordinates are output, and positioning confidence can be fed back according to a state covariance matrix trace. According to the method, the defect that a single tracking method is insufficient in precision or serious in drifting is overcome, the multi-source observation fusion difficulty is solved, tracking continuity and positioning precision are considered, accumulated errors are restrained, stable and reliable real space positioning can be achieved, the method adapts to complex working conditions, and robustness and real-time performance are better.
Need to check novelty before this filing date? Find Prior Art