Object Awareness for Intelligent Vehicle Systems: A Software Framework for Selective and Image-Aligned Presentation of YOLO Perception Outputs

PRISM (Perception-to-Risk Interface for Situational Mediation) couples object detection, monocular geometry, heuristic priority scoring, and bounded image-aligned cues to investigate how vehicle perception outputs can be translated into selective visual information. The framework combines cue-count and representation-cost limits with temporal persistence, adaptive detail, and spatial attenuation to control display activity. This software study evaluates allocation consistency using 554 COCO images, 2580 synthetic frames from eight scenes, 360 allocation configurations, analytic stress tests, and a 1200-frame real-video demonstration. Against an unthresholded risk ranking (risk budget), clear-scene PRISM reduces cue count by 54.7%, switching by 81.6%, and painted area by 30.1%, while displaying 213 of 325 score-defined reference frames, including 213 of 227 detected references. Simpler dwell ranking retains the same references with fewer transitions, identifying persistence as a major contributor to display stability. A paired geometry-to-selection diagnostic shows that box-height noise changes selected identities in 12.3% or 64.3% of observations with one or five cue slots, respectively, under specified competing-candidate conditions. Logged state and allocation time averages 0.117 ms per frame, excluding decoding, rendering, and display. The real recording demonstrates executable integration and false-cue propagation. Together, these results characterize the trade-offs among information retention, display activity, and geometric sensitivity, supported by publicly available code, experimental records, and reproducible analyses.