摘要
针对现有的盲道检测算法容易受到光照、阴影等影响,导致分割效果差的问题,提出基于改进Mask RCNN的盲道检测算法。为了提高网络的检测能力,本文增加一个滑动窗口来增大感受野的面积。在筛选时采用软非极大值抑制算法代替非极大值抑制算法,减少了目标的漏检和误检等问题。最后在深度学习框架中经过多次迭代训练,得到优化的检测模型。复杂场景下的实际测试结果表明,该算法适用于多种场景下的盲道井盖检测,具有较好的检测效果。
In order to solve the problem that the existing tactile paving detection algorithm is easily affected by light,shadow,etc.,which leads to poor segmentation effect,a tactile paving detection algorithm based on improved Mask RCNN is proposed.In order to improve the detection capability of the network,this article adds a sliding window to increase the area of the receptive field.The use of soft non maximum suppression algorithm instead of non maximum suppression algorithm during screening reduces issues such as missed and false detections of targets.Finally,after multiple iterations of training in the deep learning framework,an optimized detection model was obtained.The actual test results in complex scenes show that the algorithm is suitable for tactile paving manhole cover detection in various scenes,and has a good detection effect.
作者
黄宁霞
朱亮
HUANG Ningxia;ZHU Liang(Zunyi Vocational and Technical College,Guizhou Zunyi 563000;Guizhou Normal University,Guiyang 550000,Guizhou)
出处
《长江信息通信》
2025年第1期39-42,共4页
Changjiang Information & Communications
