摘要
针对竞赛系统机器人运动目标检测存在无用区域,易出现噪声以及空洞现象,存在误检率较高的问题,提出了基于深度学习的竞赛系统中机器人运动目标检测。通过机器人摄像头特点,建立运动目标图像获取模型,标定运动目标,根据标定结果深度分割图像提取运动目标所在区域、剔除无用区域、降低检测噪声。利用深度学习的卷积核反向传播特点,提取运动目标区域中的运动目标特征,建立竞赛系统四维向量,实时检测运动中的目标。实验结果表明,在同样存在背景干扰的前提下,对运动目标的误检率为0.95%,平均每帧检测时间为15.50 ms。
In order to solve the problem that there are useless areas,noises and holes in the detection of robot moving objects in competition system,and the detection rate is high,a robot moving target detection method based on deep learning is proposed.According to the characteristics of the robot camera,the image acquisition model of the moving target is established,and the moving target is calibrated.According to the calibration results,the image is deeply segmented to extract the region where the moving target is located,eliminate the useless area and reduce the detection noise.Based on the feature of convolution kernel back-propagation of deep learning,the feature of moving target in moving target area is extracted,and the four-dimensional vector of competition system is established to detect the moving target in real time.The experimental results show that under the same background interference,the false detection rate of moving target is 0.95%,and the average detection time of each frame is 15.50 ms.
作者
阳永清
YANG Yongqing(Department of Public Experiment Management,Changsha Normal University,Changsha 410100,China)
出处
《电子设计工程》
2021年第12期23-28,共6页
Electronic Design Engineering
基金
湖南省哲学社会科学基金项目(18YBJ10)。
关键词
深度学习
竞赛系统
机器人
分割
运动目标检测
deep learning
competition system
robot
segmentation
moving object detection
