Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.00854 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914230897213440 |
|---|---|
| author | Lodin, Igor Filatov, Sergii Filatova, Vira Filatov, Dmytro |
| author_facet | Lodin, Igor Filatov, Sergii Filatova, Vira Filatov, Dmytro |
| contents | We propose Motion-Compensated Latent Semantic Canvases (MCLSC) for visual situational awareness on resource-constrained edge devices. The core idea is to maintain persistent semantic metadata in two latent canvases - a slowly accumulating static layer and a rapidly updating dynamic layer - defined in a baseline coordinate frame stabilized from the video stream. Expensive panoptic segmentation (Mask2Former) runs asynchronously and is motion-gated: inference is triggered only when motion indicates new information, while stabilization/motion compensation preserves a consistent coordinate system for latent semantic memory. On prerecorded 480p clips, our prototype reduces segmentation calls by >30x and lowers mean end-to-end processing time by >20x compared to naive per-frame segmentation, while maintaining coherent static/dynamic semantic overlays. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_00854 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Motion-Compensated Latent Semantic Canvases for Visual Situational Awareness on Edge Lodin, Igor Filatov, Sergii Filatova, Vira Filatov, Dmytro Computer Vision and Pattern Recognition We propose Motion-Compensated Latent Semantic Canvases (MCLSC) for visual situational awareness on resource-constrained edge devices. The core idea is to maintain persistent semantic metadata in two latent canvases - a slowly accumulating static layer and a rapidly updating dynamic layer - defined in a baseline coordinate frame stabilized from the video stream. Expensive panoptic segmentation (Mask2Former) runs asynchronously and is motion-gated: inference is triggered only when motion indicates new information, while stabilization/motion compensation preserves a consistent coordinate system for latent semantic memory. On prerecorded 480p clips, our prototype reduces segmentation calls by >30x and lowers mean end-to-end processing time by >20x compared to naive per-frame segmentation, while maintaining coherent static/dynamic semantic overlays. |
| title | Motion-Compensated Latent Semantic Canvases for Visual Situational Awareness on Edge |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.00854 |