The incorporation of both virtual content and real world in augmented reality (AR) allows real-time engagement with the virtual objects. The selection of an appropriate tracking algorithm is important to optimise the performance of mobile AR applications given the limited processing capabilities and memories of mobile devices like smartphones. Tracking in AR consists of four essential components, namely detector, descriptor, matcher, and pose estimator. Since a descriptor substantially affects the overall performance of a mobile AR application, it must have short computational time and remains invariant to scale, rotation, and lighting changes. Studies have proposed Fast Retina Keypoint (FREAK) descriptor as the most suitable descriptor for mobile AR applications. Unlike other greyscale descriptors, FREAK has shorter computational time and is less likely to be affected by scale and rotation changes. However, it overlooks the vital colour space information. Focusing on enhancing the efficiency and robustness of FREAK, this study proposed the use of CRH-FREAK (RGB + HSV) descriptor and applied the vertical concatenation technique that combined all extracted keypoints vertically. The robustness of the proposed descriptors against scale, rotation, and lighting changes was verified using Mikolajczyk and Amsterdam Library of Object Images (ALOI) datasets. The developed CRH-FREAK descriptors used six colour spaces to describe the keypoints, which made them slower than the original FREAK. However, the size reduction of CRH-FREAK from 512 bits to 128 bits in this study successfully reduced the computational time to 29.49 ms, which was found comparable to the original FREAK. The improved efficiency and robustness of a 128-bit CRH-FREAK descriptor benefit the future development of mobile AR applications that remain invariant to scale, rotation, and lighting changes.

在增强现实（AR）中结合了虚拟内容和现实世界，可以与虚拟对象进行实时互动。鉴于智能手机等移动设备的处理能力和存储空间有限，选择合适的跟踪算法对于优化移动AR应用程序的性能非常重要。AR中的跟踪由四个基本组件组成，即检测器，描述符，匹配器和姿势估计器。由于描述符实质上影响了移动AR应用程序的整体性能，因此描述符必须具有较短的计算时间，并且对于缩放，旋转和照明变化保持不变。研究提出了快速视网膜关键点（FREAK）描述符作为移动AR应用程序的最合适描述符。与其他灰度描述符不同，FREAK的计算时间较短，并且受比例和旋转变化的影响较小。但是，它忽略了重要的色彩空间信息。为了提高FREAK的效率和鲁棒性，本研究提出了使用CRH-FREAK（RGB + HSV）描述符的方法，并应用了将所有提取的关键点垂直组合的垂直连接技术。使用Mikolajczyk和Amsterdam Object of Object Images（ALOI）数据集验证了提出的描述符针对比例，旋转和光照变化的鲁棒性。开发的CRH-FREAK描述符使用六个色彩空间来描述关键点，这使其比原始FREAK慢。但是，在这项研究中，CRH-FREAK的大小从512位减少到128位成功地将计算时间减少到29.49 ms，被发现与原始的FREAK相当。128位CRH-FREAK描述符的提高的效率和鲁棒性有益于移动AR应用程序的未来开发，该应用程序在缩放，旋转和照明变化方面保持不变。