[英]How to build an image processing model with new dataset using machine learning (Tensorflow)?
I want to extract the data shown on an instrument panel cluster of a bike using image processing and ML (mostly Tensorflow). 我想使用图像处理和ML(主要是Tensorflow)提取自行车仪表盘群集上显示的数据。 The sample of such cluster is 这样的簇的样本是
. 。
My input will be such complete image and I want to get the data such as speed shown by Speedometer, status of Neutral indicator(ON/OFF), Odometer reading etc from the cluster. 我的输入将是完整的图像,我想从集群中获取诸如速度计显示的速度,空挡指示器状态(开/关),里程表读数等数据。 Can anyone tell me what is the complete process for doing so? 谁能告诉我这样做的完整过程是什么? Or where are tutorials for the same? 还是相同的教程在哪里? I am new to ML and there are lot of tutorials on the net but I dont know which one of them is useful for me. 我是ML新手,网上有很多教程,但我不知道其中哪一个对我有用。
There are a few things to consider here. 这里有几件事情要考虑。 You will first want to process the image a little before feeding it into a neural network for instance. 例如,您首先需要对图像进行一些处理,然后再将其输入到神经网络中。 Maybe crop out the section which contains the speedometer reading and perform some kind of OCR (Optical Character Recognition) on it for the speed reading. 也许裁剪掉包含速度计读数的部分,并在其上执行某种OCR(光学字符识别)以进行速度读取。
The neutral indicator is a little easier actually, since there are only two states you can just look at a few sample images yourself and see which pixels actually change colors and write a quick check yourself, maybe something like so 实际上,中性指示器要容易一些,因为只有两种状态,您可以自己查看一些样本图像,看看哪些像素实际上会改变颜色并自己进行快速检查,也许像这样
(ON if pixel at (x,y) is green else OFF)
Do the same for the odometer, just identify the area where the reading is, crop it out and perform OCR on the smaller image. 对里程表执行相同的操作,只需确定读数所在的区域,将其裁剪并在较小的图像上执行OCR。 So now really you're just looking for OCR tutorials using TensorFlow, if that's a constraint. 因此,如果确实有限制,那么现在您实际上只是在寻找使用TensorFlow的OCR教程。
EDIT: 编辑:
you can do all of this cropping and editing the image programatically in python using a library called cv2 which you can install with pip 您可以使用称为cv2的库在python中以编程方式进行所有裁剪和编辑图像,可以通过pip进行安装
pip install opencv-python
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.