Distance Estimation for Precise Object Recognition Considering Geometric Distortion of Wide-angle Lens
-
摘要: 人脸识别和车牌识别是智慧安防领域的重要内容。人脸和车牌的特征尺度小、细节丰富，对成像系统的空间分辨力有较高要求，需要较大规模的探测器和高传函、小畸变的光学镜头。然而，安防系统又要求广域监控，需要使用视场大但具有一定畸变的广角镜头。因此，设计既能精确识别人脸和车牌目标、又能广域监控的成像系统时，应将精确目标识别作为约束来权衡高空间分辨力和大视场的性能指标以及估计识别距离。在这样的应用需求下，本文提出了像素面密度对精确目标进行统一描述，并提出了考虑广角镜头径向畸变的精确目标识别距离估算方法，通过对存在旋转和平移的人脸和车牌目标进行计算验证，结果表明：考虑径向畸变后实际识别距离较理论识别距离近，且人脸和车牌平移距离分别为1 m和2 m时，实际与理论的识别距离差异高达34.2%和27.5%。Abstract: Face and license plate recognition are crucial aspects in the field of intelligent security. A high-spatial-resolution imaging system with a large-format detector and low-distortion optical lens is required for recognizing small-scale features and rich details in faces and license plates. However, security systems need to monitor wide area, which requires a wide-angle lens with a wide field of view, but with some distortion. Therefore, precise target recognition should be used as a constraint to balance the high spatial resolution and wide field of view when designing an imaging system that can recognize details and monitor a wide area. Under such application requirements, an evaluation index based on pixel areal density is proposed. With the aid of this evaluation index, a distance estimation method for precise object recognition, considering the radial distortion of the wide-angle lens, was designed. Rotated and translated faces and license plates were used to demonstrate the estimation method. The results indicate that the recognition distance with radial distortion is less than that without radial distortion. When the translation distance is 1 m and 2 m, the difference between the actual recognition distance and the ideal recognition distance is 34.2% and 27.5%, respectively.
-
Key words:
- recognition distance /
- wide-angle lens /
- radial distortion /
- object recognition /
- pixel areal density
1) 前言:1) 红外光学系统是红外光电设备的眼睛，位于红外光电设备的最前端，是红外光电设备最重要的组成部分之一，其性能的好坏直接决定着红外光电设备的整机性能。由于红外光学系统具有被动成像、不易被干扰、识别伪装能力强、可全天时工作、探测灵敏度高等优点，因此广泛应用于光电侦察、航空航天、空间遥感、精确制导、火灾搜救、电力巡检、医学检查等领域。1) 最后，感谢专栏论文所有作者的卓越贡献。1) ——白瑜1) 为了促进科研人员在红外光学系统领域交流的最新成果，2021年12期，《红外技术》推出了“红外光学系统”专栏，共收录8篇学术论文，内容涉及二次成像结构的中波红外折反射式光学系统设计，大相对孔径的长波红外变焦无热化光学系统设计，制冷型中/长红外双波段一体化全反射式光学系统设计，考虑广角镜头畸变的精确目标识别距离估算，机载小型化中波红外连续变焦光学系统设计，红外探测器集成光学系统低温评价方法研究等，涉及领域较广，旨在集中反映报道红外光学系统领域的新动态和发展趋势，为相关科研人员和广大读者提供学术价值参考，为红外光学系统的研究发展提供一些新的思路。1) 红外光学系统涉及光学理论，像差理论，光机优化，光学材料，红外器件，光学制造，集成测试等多个领域，随着科学技术的快速发展，红外光学系统向着多次成像结构，多视场，大相对孔径，宽温度范围，多波段，一体化，小型化，集成化，轻量化等方向发展。 -
图 3 识别距离与旋转角度的关系
Figure 3. Relationship between recognition distance and rotation angle
图 4 像素面密度与平移距离的关系
Figure 4. Relationship between pixel areal density and translation distance
表 1 识别标准与识别指数
Table 1. Standards and recognition index
Standard Evaluation index Identification China (Face) Pixels ≥30 China (License plate) Pixels ≥100 European Union (Person) mm/pixel > 4 United Kingdom (Person) TSR > 100% 
下载: 导出CSV
表 2 识别人脸、车牌和人所需的最小像素面密度
Table 2. Minimal pixel density of face, license plate and person recognition
Situations Pixel density(pixels/m2) Face recognition 267289 License plate recognition 51984 Person detection 625 Person recognition 15625 Person identification 62500
下载: 导出CSV
表 3 识别距离与视场角和分辨率的关系
Table 3. Relationship between recognition distance and field of view, recognition distance and resolution
Resolution (pixel) Recognition distance of face/License plate/m 50° 60° 70° 80° 90° 100° 110° 1280×720 2.66/6.02 2.15/4.87 1.77/4.01 1.48/3.35 1.24/2.81 1.04/2.36 0.87/1.97 1920×1080 3.99/9.03 3.22/7.30 2.66/6.02 2.22/5.02 1.86/4.22 1.56/3.54 1.31/2.95 2560×1440 5.31/12.04 4.29/9.73 3.54/8.02 2.96/6.70 2.48/5.62 2.08/4.72 1.74/3.94 3840×2160 7.97/18.06 6.44/14.59 5.31/12.03 4.43/10.04 3.72/8.43 3.12/7.07 2.61/5.90
下载: 导出CSV
表 4 实际识别距离与理论识别距离的差异
Table 4. Difference between actual recognition distance and ideal recognition distance
Translation distance /m Ideal recognition distance/m Actual recognition distance/m Difference in distances/m Difference in proportions Face recognition 0.25 2.22 2.18 0.04 1.8% 0.5 2.22 2.07 0.15 6.8% 0.75 2.22 1.82 0.40 18.0% 1.0 2.22 1.46 0.76 34.2% License plate recognition 0.5 5.02 4.96 0.06 1.2% 1.0 5.02 4.76 0.26 5.2% 1.5 5.02 4.37 0.65 12.9% 2.0 5.02 3.64 1.38 27.5%
下载: 导出CSV
-
[1] Taskiran M, Kahraman N, Erdem C E. Face recognition: past, present and future (a review)[J]. Digital Signal Processing, 2020, 106: 102809. doi: 10.1016/j.dsp.2020.102809 [2] DING C, TAO D. A comprehensive survey on pose-invariant face recognition[J]. Acm Transactions on Intelligent Systems & Technology, 2016, 7(3): 37. http://www.researchgate.net/profile/Changxing_Ding/publication/272422850_A_Comprehensive_Survey_on_Pose-Invariant_Face_Recognition/links/54f9177b0cf28d6deca34081.pdf [3] WANG Zhenzhou, ZHANG Xu, YU Pingping, et al. A new face recognition method for intelligent security[J]. Applied Sciences, 2020, 10(3): 852. doi: 10.3390/app10030852 [4] Ahmed A U, Masum T M, Rahman M M. Design of an automated secure garage system using license plate recognition technique[J]. International Journal of Intelligent Systems & Applications, 2014, 6(2): 22-28. [5] Baran R, Rusc T, Fornalski P. A smart camera for the surveillance of vehicles in intelligent transportation systems[J]. Multimedia Tools and Applications, 2016, 75(17): 10471-10493. doi: 10.1007/s11042-015-3151-y [6] CHUNG Y N, LU T C, HUANG Y X, et al. Applying image recognition technology to intelligent security management systems[C]//International Conference on Artificial Intelligence and Industrial Engineering (AIIE 2015), 2015: 117-120. [7] Abramkina D E, Khatsevich T N. Analysis of optical systems of extremely wide-angle lenses for digital cameras[J]. Interexpo GEO-Siberia, 2020, 6(1): 3-10. doi: 10.33764/2618-981X-2020-6-1-3-10 [8] Juarez-Salazar R, ZHENG J, Diaz-Ramirez V H. Distorted pinhole camera modeling and calibration[J]. Applied Optics, 2020, 59(36): 11310-11318. doi: 10.1364/AO.412159 [9] WU F, WEI H, WANG X. Correction of image radial distortion based on division model[J]. Optical Engineering, 2017, 56(1): 013108. doi: 10.1117/1.OE.56.1.013108 [10] Johnson J. Analysis of image forming systems[C]//Proceedings of SPIE, 1958, 513(513): 761. [11] Leachtenauer J C. Resolution requirements and the Johnson criteria revisited[J]. Proceedings of SPIE, 2003, 5076: 1-15. doi: 10.1117/12.497896 [12] 中国国家标准化管理委员会. 公共安全人脸识别应用图像技术要求: GB/T 35678-2017[S]. 中国标准出版社, 2017.Standardization Administration of the People's Republic of China. Public Security Face Recognition Application Technical Requirements for Face Images: GB/T 35678-2017[S]. Standards Press of China, 2017. [13] 中国国家标准化管理委员会. 机动车号牌图像自动识别技术规范: GA/T 833-2016[S]. 中国标准出版社, 2016.Standardization Administration of the People's Republic of China. License plate of motor vehicle of the People's Republic of China: GA/T 833-2016[S]. Standards Press of China, 2016. [14] European Committee for Electrotechnical Standardization. Video surveillance Systems for Use in Security Applications-Part 4: Application Guidelines: UNE-EN 62676-4-2015[S]. CEN-CENELEC Management Centre, 2014. [15] Tony Mansfield. Adding Value to CCTV: Issues in Testing of Facial Recognition Systems[M]. UK: Home Office Scientific Development Branch, 2014: 4-12. [16] 中国国家标准化管理委员会. 成年人头面部尺寸: GB/T 2428-1998[S]. 中国标准出版社, 1998.Standardization Administration. Head-face Dimensions of Adults: GB/T 2428-1998[S]. Standards Press of China, 1998. [17] 中国国家标准化管理委员会. 中华人民共和国机动车号牌: GA 36-2018[S]. 中国标准出版社, 2018.Standardization Administration. License Plate of Motor Vehicle of the People's Republic of China: GA 36-2018[S]. Standards Press of China, 2018. -
点击查看大图
图(6) / 表(4)
计量
- 文章访问数: 205
- HTML全文浏览量: 81
- PDF下载量: 63
- 被引次数: 0
