Dhwanil R. Chauhan

We present the first unified framework for novel-view acoustic synthesis that entirely bypasses explicit 3D visual rendering and costly photogrammetry by directly grounding spatial audio generation in feed-forward visual geometry. Our framework synthesizes accurate and immersive spatial audio in 3D spaces without requiring viewpoint images, dense point maps, or ground-truth poses for input video. We propose the Geometry-Grounded Acoustic Decoder (GGAD) to dynamically attend to cross-modal features embedding local and global geometries in audio and visual modalities. Extensive experiments show that our framework outperforms prior work across various benchmarks in high-quality, viewpoint-accurate spatial audio synthesis, without requiring time-consuming explicit rendering of novel-view images or dense point maps.

We present a multi-camera perception system for dynamic safety zone reconfiguration in active industrial environments. Static safety zone definitions fail in dynamic industrial floors where equipment and personnel positions shift continuously. Our system fuses four simultaneous camera perspectives using a rule-based spatial reasoning engine to redefine green zone boundaries in real time based on outermost detected blocker positions, enabling proactive hazard detection before incidents occur.

The robustness of vision-language models under simultaneous visual and linguistic degradation — a common failure condition in real-world deployments — remains largely unstudied. We introduce a systematic benchmark evaluating 20 VLMs across controlled simultaneous multimodal corruption conditions, featuring a novel text corruption module and structured evaluation pipeline designed to characterize failure modes relevant to safety-critical and industrial deployment scenarios.

Experimental evaluation demonstrates strong performance, achieving an F1 score of 0.92 for worker detection and 0.93 for personal protective equipment (PPE) classification. The integrated hazard-zone monitoring module further enables spatial safety enforcement, achieving a frame-level detection accuracy of 92% for restricted-area violations. The system supports real-time monitoring and processes video streams at approximately 35 ms per frame on a standard GPU. Preliminary deployment observations in an operational steel facility suggest that the system can assist safety personnel by improving visibility of PPE compliance and restricted-area monitoring. These results demonstrate the technical feasibility of the proposed modular framework as a scalable approach for automated industrial safety monitoring.