光子学报, 2019, 48 (2): 0212003, 网络出版: 2019-03-23
仿复眼视觉测距传感器视轴的快速标定法
Rapid Optic Axis Calibration of Bio-vision Ranging Sensors Based on Compound Eyes
生物光学 多视轴标定 曲线拟合 仿生复眼 三角测量 侧抑制神经网络 运动估计 Biological optics Muti optical axis calibration Curve fitting Artificial compound eye Triangulation Lateral inhibition neural network Motion estimation
摘要
基于生物复眼的视觉结构特性及视神经处理机制, 搭建了环形31通道的仿复眼视觉快速测距传感器, 建立了基于侧抑制神经网络的仿生多通道视觉快速测距模型, 并提出了基于高斯分布离散信息的多通道传感器多视轴快速标定法, 实现环形传感器视轴标定.实验结果表明:传感器视轴标定后, 传感器阵列中间视角区左眼相对准确度提高20.46%, 右眼相对准确度提高9.00%; 采用夹角6°的传感器阵列, 在左右眼中间评测区内80%以上区域内测得目标所在角度误差小于±0.60°; 实现了仿复眼视觉测距传感器对运动目标(速度50 mm/s)定位误差在-3~10 mm间的实时测距.
Abstract
Based on the processing mechanism of optic nerve and the characteristics of visual structure of biological compound eye, a ring-shaped vision sensor based on bionic compound eye with fast distance measurement function and 31 channels is built, and a bionic multi-channel fast vision distance measurement model based on lateral inhibition neural network is established. To calibrate optic axis of ring sensors, a rapid way for optic axis calibration of muti-channel sensors based on discrete information with Gaussian distribution is proposed. The experimental results show that the relative accuracy of the left eye is increased by 20.46% and that of the right eye is increased by 9.00% in intermediate angle of view of sensor array after the calibration of optic axis of sensors. The angle measurement accuracy of more than 80% of the points in the middle evaluation area of the left and right eye is within ±0.60° with the sensor array that use 6° angle spacing. The location error of binocular real time ranging results is kept within the range of -3 mm to 10 mm under 50 mm/s velocity.
杨预立, 邢强, 姚建南, 戴振东, 王国军, 徐海黎. 仿复眼视觉测距传感器视轴的快速标定法[J]. 光子学报, 2019, 48(2): 0212003. YANG Yu-li, XING Qiang, YAO Jian-nan, DAI Zhen-dong, WANG Guo-jun, XU Hai-li. Rapid Optic Axis Calibration of Bio-vision Ranging Sensors Based on Compound Eyes[J]. ACTA PHOTONICA SINICA, 2019, 48(2): 0212003.