MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641109032 A) filed by Easwari Engineering College; and Srm Institute Of Science And Technology,ramapuram Campus on September 11, 2026, for Vision-Language-Based Cross-View Attribute Fusion For Fine-Grained Person Re-Identification.
Inventors include Nithish Khanna S; Hari Prakash G; and Saranya D.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: ABSTRACT OF THE INVENTION: The present invention relates to a computer-implemented system and method for vision-language-based cross-view attribute fusion designed to enhance fine-grained person re-identification (Re-ID). Existing vision-language Re-ID systems treat textual descriptions as independent per-image representations, failing to resolve attribute discrepancies caused by viewpoint changes or environmental occlusions. To address this fundamental limitation, the proposed system integrates a novel analytical layer downstream of standard multimodal retrieval. This layer compares structured pedestrian attributes—extracted via domain-specific prompt chaining utilizing a local Vision-Language Model—across diverse, non-overlapping camera views. The analysis module explicitly classifies each compared attribute slot as persistent, potentially variable, view-dependent/occluded, or definitively contradictory based on natural variability priors. It then fuses these classifications into an interpretable, evidence-aware Identity Evidence Score, mathematically discounting contradiction penalties based on attribute-specific occlusion likelihoods. By effectively distinguishing genuine identity conflicts from unobserved features, the framework significantly reduces the manual reconciliation burden in human-inthe-loop retrieval and provides a structured, human-readable evidence summary to support transparent candidate verification across vast surveillance networks. Total Number of Words in Abstract: 164
Disclaimer: Curated by HT Syndication.
