研究目的
To determine the optimum NIR filters based on the stability of the camera calibration parameters and to evaluate the accuracy of mapping for surveillance applications in low-light conditions.
研究成果
The 850 nm NIR filter was identified as the most optimum for surveillance applications due to its stability in camera calibration parameters and high accuracy in 3D measurements, with differences less than ±0.8 mm compared to true values. This filter produced clearer and sharper images in low-light conditions, making it suitable for forensic mapping and suspect identification. Future studies should explore NIR emitter characteristics like beam-width and strength to enhance performance.
研究不足
The study used a mannequin instead of a human subject, which may not fully replicate real-world scenarios. The camera calibration was limited to certain parameters (e.g., only K1 for lens distortion), and the environment was controlled, potentially not accounting for all external variables in surveillance applications. Further optimization could involve testing with actual humans and varying environmental conditions.
1：Experimental Design and Method Selection:
The study involved geometric calibration of a camera with and without NIR filters, followed by a case study for 3D measurement. A high-precision calibration frame with retro-targets was used, and convergent photographs were captured from multiple stations. Statistical ANOVA was employed to analyze differences in camera parameters and 3D measurement accuracy.
2：Sample Selection and Data Sources:
A mannequin was used as the subject to avoid movement errors, with test points measured using a Microscribe for true values. Data were collected from photographs under normal and low-light conditions with different NIR filters.
3：List of Experimental Equipment and Materials:
Sony HDR-HC5E HDV camera, NIR filters (715nm, 780nm, 830nm, 850nm), Sony HVL-IRM Infrared Night Shot Light illuminator, calibration frame with rods and retro-targets, invar scale bars, tripod, and software (Australis for photogrammetry, MegaStat for statistical analysis).
4：Experimental Procedures and Operational Workflow:
For calibration, eight convergent photographs were captured from four positions with landscape and portrait orientations. For the case study, photographs of the mannequin were taken with each filter type, repeated three times, in low-light conditions. Images were digitized and processed using Australis software to compute 3D coordinates and distances.
5：Data Analysis Methods:
ANOVA was used to test for significant differences in camera parameters (focal length, principal point coordinates, lens distortion) and 3D measurement accuracy. Standard deviations and mean differences were calculated to evaluate performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
