Wavefront-Constrained Passive Obscured Object Detection

TL;DR

This work tackles passive obscured object detection in non-line-of-sight scenarios by embedding physics-informed complex-wave propagation into a neural framework. The proposed TriWCP module confines optimization to a physically admissible, low-rank complex subspace, while the Cross-layer Compensation enhances high-frequency details through multi-scale frequency filtering and semantic alignment. Together, these components deliver robust localization and segmentation of occluded targets under extreme low-SNR conditions, with strong empirical gains across four physically collected datasets and comprehensive ablations. The approach offers a practical, hardware-friendly pathway for reliable NLOS sensing in real-world settings.

Abstract

Accurately localizing and segmenting obscured objects from faint light patterns beyond the field of view is highly challenging due to multiple scattering and medium-induced perturbations. Most existing methods, based on real-valued modeling or local convolutional operations, are inadequate for capturing the underlying physics of coherent light propagation. Moreover, under low signal-to-noise conditions, these methods often converge to non-physical solutions, severely compromising the stability and reliability of the observation. To address these challenges, we propose a novel physics-driven Wavefront Propagating Compensation Network (WavePCNet) to simulate wavefront propagation and enhance the perception of obscured objects. This WavePCNet integrates the Tri-Phase Wavefront Complex-Propagation Reprojection (TriWCP) to incorporate complex amplitude transfer operators to precisely constrain coherent propagation behavior, along with a momentum memory mechanism to effectively suppress the accumulation of perturbations. Additionally, a High-frequency Cross-layer Compensation Enhancement is introduced to construct frequency-selective pathways with multi-scale receptive fields and dynamically model structural consistency across layers, further boosting the model's robustness and interpretability under complex environmental conditions. Extensive experiments conducted on four physically collected datasets demonstrate that WavePCNet consistently outperforms state-of-the-art methods across both accuracy and robustness.

Figures (4)

• Figure 1: Typical scenario underlying the passive OOD regime, in which the target, entirely occluded by an opaque intervening barrier that prevents any direct radiative conveyance to the sensor, is inferred solely through the diffuse, low-intensity speckle distribution scattered onto the visible relay surface.
• Figure 2: Schematic representation of WavePCNet architecture, wherein a bifurcated topology integrates (i) a physically constrained optimisation pipeline grounded in complex-amplitude propagation (TriWCP), enforcing low-rank spectral admissibility via manifold-constrained trajectory reprojection; and (ii) a cross-layer compensatory stream that augments attenuated high-frequency constituents through semantics-guided, frequency-selective enhancement.
• Figure 3: Architectural composition of the proposed modules. Left: Cross-layer Compensation integrates multi-receptive field frequency-selective filtering and semantics-guided structural alignment to recover attenuated high-frequency cues. Center: spatial-frequency attention projection across layers enhances structural consistency. Right: Tri-Phase Complex-Propagation Reprojection constrains optimisation within a physically admissible low-rank subspace via embedded Fresnel transfer operator and memory-augmented updates.
• Figure 4: Visual comparison of detection results across representative methods on occluded-target datasets.