Forensic Detection System Using Video-Stabilized Coded Aperture Snapshot Spectral Camera
Sehoon Lim, Choongyeun Cho, and Sek Chai
IEEE International Conference on Technologies for Homeland Security 2015 (Accepted), Apr 2015
Abstract
Multispectral imaging (MSI) enables the capture of the spectral properties of materials in a scene based upon their spectral responses to current illumination (e.g. fluorescence, reflection, absorption). We describe several key developments towards the realization of a small form-factor MSI instrument capable of instantaneous capture and analysis of the spectral signatures of all objects in the scene. A multi-frame Coded Aperture Snapshot Spectral Imaging (CASSI) is proposed by Video Stabilization (VS). Imperfect recovery of spectral cube is minimized by using the stabilized multi-frames in video imaging. The CASSI system optically encodes a spectral cube and computationally decodes the cube, which enables video-rate detection. An Adaptive Cosine Estimator (ACE) is used to quantitatively score target objects in the scene from the decoded spectral cube. This system thus enables spatial localization of target object by using a highly robust ACE spectral analysis within the processing flow. In this paper, we present a demonstration of VS-CASSI with improved results indicating its detection and classification performance.There are a number of important applications for the proposed CASSI-based MSI technology ranging from forensic evidence to security checkpoint screening. Its ability to provide snapshot spectral datacube instantaneously allows for fast data capture even with camera motion. The proposed architecture is modular and configurable to support imaging in different spectral bands and with varying degree of accuracy in spectral analysis. Different requirements such as capture time, sensitivity to sample size and standoff distances can be considered in the design while maintaining the same architecture.
For forensic evidence detection, the proposed system offers the user the ability to quickly scan an area, create a mosaic of spectral information of the scene, and determine which samples to gather, or which area require further analysis. At present, forensic analysts illuminate darkened forensic scenes using spectrally selective light sources while searching for evidence through various pairs of colored glasses, just as they have been doing for the past 30 years. The techniques currently used for this often result in large quantities of non-probative samples being collected for laboratory analysis, making the screening process both overly expensive and time-consuming.
It is envisioned that the proposed handheld MSI camera could be programmed to highlight specific evidential materials such that an investigator could simply capture and detect them at forensic scene areas of interest, whereupon supported selected sample type would be indicated on a large touchscreen display integrated into the instrument. The samples thus detected could then be quickly located and collected with high confidence regarding their forensic importance; their type and anticipated probative value could also be simultaneously stored along with any other relevant scene metadata. For operation in the 300nm to 550nm band (Near Ultraviolet to mid-visible), those evidential materials with significantly unique spectral emission signatures such that a VS-CASSI + ACE system could be readily programmed to detect them include: semen, blood, saliva, hair, vaginal fluid, fingerprints (skin oil), cocaine and methamphetamine. For other spectral regions (e.g. visible, near infrared, shortwave infrared), further materials including nitrate based explosives ould be detected, even in trace quantities. Using such an instrument, more truly probative and less non-probative samples could ultimately be collected, thus minimizing costly lab time while enabling the timely processing of relevant evidence. Wet or dry, visible or not, this detection implementation could ensure precise evidence collection on virtually any substrate material, fluorescent or not.