【摘要】 针对当前视觉即时定位与地图构建(VSLAM)前端视觉惯性里程计(VIO)存在运算量大导致计算时间长的问题，提出了一种全新的基于图形处理器(GPU)并行加速的VIO方法。首先，对VIO进行加速算法设计，包括限制对比度的自适应性直方图均衡化(CLAHE)算法加速、FAST角点筛选改进加速以及改进光流跟踪算法加速。最后，将设计的加速算法串并组合成前端，结合开源VINS-Mono后端进行定位精度与实时性测试，平均耗时减少12.03 ms,定位精度均方根(RMS)值相差0.008 963 m。实验结果表明：提出的方法在保持定位精度的同时提升了算法实时性能，且优于VINS-Mono中基于OpenCV GPU加速的方法。更多还原

【Abstract】 Aiming at the problem of long calculation time caused by a large amount of calculation of current visual inertial odometer(VIO) of the front-end visual simultaneous localization and mapping(VSLAM),and a new VIO method based on graphics processing unit(GPU)parallel acceleration is proposed.Firstly, VIO is designed to accelerate the algorithm, including contrast limited adaptive histogram equalization(CLAHE)algorithm acceleration, FAST corner screening improved acceleration, and improved optical flow tracking algorithm acceleration.Finally, the designed acceleration algorithm is combined into a front-end in series and parallel, combined with the open source VINS-Mono back-end for positioning precision and real-time testing, the average time consumption is reduced by 12.03 ms, and the positioning precision RMS difference is 0.008 963 m.Experimental results show that the proposed method improves the real-time performance of the algorithm while maintaining the positioning precision, and it is prior to the method based on OpenCV GPU acceleration in VINS-Mono.更多还原