[Paper Review] Tracking moving objects through scattering media via speckle correlations

Y. Jauregui-Sánchez, Harry Penketh+1 more|arXiv (Cornell University)|Feb 22, 2022

Random lasers and scattering media4 citations

TL;DR

This paper proposes a real-time, low-complexity method to track moving objects behind scattering media using cross-correlations of time-varying speckle patterns. By computing the difference between cross-correlations and autocorrelations of speckle intensities at consecutive frames, the technique isolates motion signals, enabling tracking beyond the optical memory effect range with minimal computation and no image reconstruction.

ABSTRACT

Scattering can rapidly degrade our ability to form an optical image, to the point where only speckle-like patterns can be measured. Truly non-invasive imaging through a strongly scattering obstacle is difficult, and usually reliant on a computationally intensive numerical reconstruction. In this work we show that, by combining the cross-correlations of the measured speckle pattern at different times, it is possible to track a moving object with minimal computational effort and over a large field of view.

Motivation & Objective

• Address the challenge of tracking moving objects behind strongly scattering media where conventional imaging fails due to speckle degradation.
• Overcome the computational burden of reconstructing images through scattering media by focusing on motion tracking instead of full image recovery.
• Enable real-time tracking with minimal computational cost by exploiting temporal correlations in speckle patterns.
• Extend the effective field of view beyond the optical memory effect range by leveraging background speckle correlations.
• Develop a method that is robust to complex scenes with static backgrounds by subtracting static components from cross-correlations.

Proposed method

• Measure time-series speckle intensity patterns from a scene hidden behind a scattering medium using a lensless CMOS camera.
• Compute cross-correlations between speckle patterns at consecutive time steps to extract motion-related information.
• Apply the difference between cross-correlations and autocorrelations to suppress static background contributions.
• Use the resulting signal, which has zero mean and encodes motion direction, to generate interpretable motion frames.
• Leverage the fact that only speckle patterns correlated within the memory effect range contribute to the signal, enabling tracking even when the object moves outside the standard memory range.
• Utilize a dual-layer adhesive tape to reduce the memory effect range, thereby enhancing sensitivity to local motion while preserving transmission.

Experimental results

Research questions

• RQ1 Can motion of an object hidden behind a strongly scattering medium be tracked without full image reconstruction?
• RQ2 Can cross-correlations of time-varying speckle patterns reveal motion information with minimal computational cost?
• RQ3 Is it possible to extend the effective field of view for tracking beyond the optical memory effect range?
• RQ4 How can static background components be effectively suppressed in speckle-based motion tracking?
• RQ5 Can the method detect not only translational motion but also rotation, deformation, or size changes of hidden objects?

Key findings

• The proposed method enables real-time tracking of moving objects behind scattering media with negligible computational overhead.
• By subtracting the autocorrelation from the cross-correlation, static background components are effectively canceled, leaving only motion-induced signals.
• The technique successfully tracks motion even when the object moves beyond the optical memory effect range, provided it remains within correlation range of at least one static background point.
• The resulting motion signal is centered on the moving object, with the background appearing to move relative to it, enabling intuitive interpretation of motion direction.
• Experimental results using a DMD-based scene and two layers of adhesive tape confirmed that motion tracking is robust and effective even with reduced memory effect range.
• The method is applicable to various motion types, including translation, rotation, and deformation, without requiring prior knowledge of object shape or motion model.