神经网络下单目视觉测距方法的研究

Study on monocular vision distance measurement based on neural network

依据高斯成像原理和自动调焦技术,提供一种由变焦镜头获得焦距与像距的双脉冲数,利用神经网络计算物距的单目视觉测距新方法.该方法根据光学成像原理,计算不同物距下对应目标图像的清晰度值,根据改进的变步长爬山搜索策略实现自动调焦,记录电机转向和对应脉冲数,再根据获得的实验数据训练神经网络实现测距.最后提出面积法确定焦距脉冲数,进一步提高测量速度与精度.以粮仓测距为例,神经网络训练结果表明,样本误差、输入与输出的回归系数和相对误差均满足精确度要求,验证了该方法的可行性.该方法避免了复杂的摄像机标定,计算简单且易于实现,为测距技术的研究提供新的参考和理论依据.

A new method of monocular vision ranging is presented based on gaussian imaging principle and automatic focusing technology in the paper.The distance is calculated by neural network and the double pulse of focal length and image distance provided by zoom lens.Firstly,the sharpness values of the target image of various object distances are obtained according to the principle of optical imaging.Secondly,automatic focusing is achieved by the improved and varied step-length climbing search strategy,meanwhile,the stepping motor steering and the number of pulse are recorded.Then,the distance measurement is accomplished by the neural network,trained based on experimental data.Finally,the area method is used to confirm the number of focal length pulse,and further improve the measurement speed and accuracy.Taking the barn ranging as an example,the neural network training result shows that the sample error and the input and output regression coefficients and the mean of relative error satisfy the accuracy requirements.The feasibility of the method is verified.The complicated camera calibration is avoided and the calculation is simple and easy to implement in this method,which provides a new reference and theoretical basis for the research of distance measurement technology.