【摘要】 场景中的动态物体影响移动机器人定位算法的精度,使机器人无法建立蕴含场景信息的高精度地图,降低定位建图系统在复杂场景中的鲁棒性。针对目前主流动态SLAM技术受限于系统需求和硬件性能,无法兼顾移动机器人定位精度和系统实时性的问题,提出一种基于YOLO v3和传感器融合的机器人定位建图系统。首先,建立融合编码器和视觉传感器的机器人运动模型,求解移动机器人位姿;然后,利用深度学习技术剔除复杂场景中的动态物体,并针对YOLO v3目标检测网络特点,采用多视图几何方法进行性能优化;最后,经测试,本系统相比DS_SLAM具有更优的轨迹精度,耗时更短。更多还原

【Abstract】 The existing thermal wave detection technology for structural adhesive damage of glass curtain wall has some problems, such as large amount of thermal image sequence data, less effective information, low resolution and large noise. The thermal image sequence of glass curtain wall is completed by using data reconstruction of single column position, image reconstruction based on wavelet transform, image enhancement based on Wiener filter and thermal wave location based on damage area recognition rule Column processing and damage area identification. The experimental results show that: the wavelet transform technology using adaptive threshold coefficient can effectively reduce the noise components in the high-frequency components of the thermal image and retain the characteristics of the thermal image; Wiener filter uses 3 × 3 template to further smooth the image to ensure most of the important information in the thermal image; the recognition rate of the damage area is 93.7%.更多还原