Engineering of License Plate Recognition System Algorithm - Tigerwong

But license plate recognition system, are you sure your user has a graphics card? In your own company or laboratory, you may not believe that a company with hundreds of it teams will not have graphics cards. The hardware conditions can be basically met; But to the user, everything may be different. You need to run it on the customer's computer / server. There are many projects. In many environments, there is no GPU, or if you do engineering, GPU may be a luxury. For the computer vision project, if the customer needs to carry the model on the unique hardware (such as the defect detection equipment in the field of Intelligent Manufacturing), it is very important to confirm the hardware configuration with the user in advance.

Whether the customer knows what GPU is is absolutely the watershed of the difficulty of cooperation. It will be easier to know and know what GPU is; If you know nothing about GPU or only use Intel GPU, the landing difficulty increases exponentially. A simple criterion: when communicating with customers. Be sure to communicate with users in advance. Graphics cards are hard to find.Do you have the data? The first question in the internal discussion must be whether the project has data? All items. There are often ready-made data sets when technicians reproduce papers. Cleaned data. Once the project is started, obtaining data sets and data set preprocessing are the most time-consuming links. After taking over many projects and discussing all the business details, the technicians found that the customer had no data or the data was not enough. With the data... Can you understand it? For most computer vision projects, the last question follows.I really can't understand the customer data of the license plate recognition system. For example, medical images and data labeling are things that doctors need years of training to do well. It is difficult for sequencers without any medical background to understand these data in a short time. At the beginning of the year, I did defect identification for a factory and figured out what network and layout mode to use. As a result, I couldn't understand the pictures and couldn't mark them.

It looks like everything is broken, but the quality inspector of the factory can distinguish between normal and defects at a glance. The knowhow of this industry is very important. It is difficult for non-medical personnel to see which angiocardiograms are helpful for diagnosis. Understand each other's expectations? It has created an impression that "artificial intelligence is omnipotent". The information asymmetry between users and technicians leads to different expectations for the project. In the face of users who have never been in contact with artificial intelligence, due to the exaggerated reports of the artificial intelligence industry in the media in recent years.

Ensure that users understand the following points: but this accuracy cannot be guaranteed. In any case that is very far from the training data, 1. Only the accuracy of the test set is guaranteed. It can be reproduced; 2. Machine learning projects cannot achieve 100% accuracy; It's not like you can use it by buying a Microsoft offline download. 3. Training takes time. The most annoying thing about engineering is to do a lot of demos for customers. You need to do a demo for any demand.These demos are a waste of time, but we can't help doing them. So in order to save time for each computer vision project, I finally. We have made tegucv a visual computer vision and neural network tool. At present, it supports image recognition, video classification, face comparison and license plate recognition, and tegucv supports the pure local operation of window. The license plate recognition system also includes automatic parameter selection, HTTP server and other functions.Start training the computer vision model in ten minutes. A lot of simple work can be realized directly with tegucv. It is convenient for developers to quickly implement the technology, and can peel away a lot of simple work.