This report presents the methodology and results of the testing and evaluation (T&E) of a second-generation (Gen2) long-range facial recognition binocular system called 3DMobileID, which was developed by Stereo Vision Imaging, Inc. (SVI) under funding by the National Institute of Justice (NIJ).
The testing was conducted by NIJ's Sensors, Surveillance, and Biometric Technologies (SSBT) Center of Excellence (CoE). The T&E involved a technology assessment of the device's functionality, a data collection of face images from 100 subjects, and an evaluation of the match performance using two industry standard algorithms. The general conclusion of the T&E is that the Gen2 system shows a dramatically improved performance over the Gen1 device; however, there are still issues that must be addressed before the device becomes an operational product for law enforcement use. It may need to include more capabilities, such as video capture, before law enforcement is willing to adopt the system. Removing the background through the occlusion process improved the match performance of the system when using the L1 matcher. The face image datasets were submitted as probe sets against the enrollment gallery using L- 1 Identity Solutions (L1). Occluded images did not improve the performance of matching runs and can sometimes hinder the performance of the system. In addition, the integration of the L1 matchers with the Disparity Calculator is incorrectly set up. It was determined that when braced on a tripod, the tripod and hand-held modes perform comparably. It is recommended that SVI and future T&E efforts conduct a parametric study of the effect of Disparity Calculator occlusion parameter on matching performance. Further improvements to the occlusion image process and subsequent match performance may be realized by exploring this parameter space. 76 figures, 39 tables, and appended supplementary information