# mlnet.onnx.openalpr

**Repository Path**: weyhd_geffzhang/mlnet.onnx.openalpr

## Basic Information

- **Project Name**: mlnet.onnx.openalpr
- **Description**: No description available
- **Primary Language**: C#
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2021-07-01
- **Last Updated**: 2024-09-01

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# ONNX Object Detection + License Plate Detection
An object + license plate detection (from a video/image) solution, leveraging Open ALPR (.NET Framework) & ML.Net ONNX Detection (.NET Core).

| Name | Repository |
| ------ | ------ |
| Open ALPR | [Github Repository](https://github.com/openalpr/openalpr) |
| ML ONNX Object Detection | [Github Repository](https://github.com/dotnet/machinelearning-samples/tree/master/samples/csharp/end-to-end-apps/ObjectDetection-Onnx) |

### Getting Started

+ The startup project should be MLNetOnnxAlpr.SampleUi [Build x64]
+ Change the file in MainWindow.xaml.cs -> var videoFilePath = String.Empty; [Line 77]

### Points of Interest

##### MS ML Onnx Object Detection

+ This is a modified version of Microsoft’s samples "OnnxObjectDetectionWPFApp" & "OnnxObjectDetection", leveraging TinyYolo2_model.onnx.
+ There were a few refactors / modifications on the overlay methods, in which it will remove overlapped overlays of identified objects.
+ I looked into adding the duc.onnx model yet was unable to get the bounding boxes correct [Duc](https://github.com/onnx/models/tree/master/vision/object_detection_segmentation/duc).

##### Open ALPR

+ The source has been in a rough state for the last few years based on the issues/comments yet the last public release was close enough to the sample to piece it together.
+ I attempted to rebuild with the latest source yet their build instructions were leveraged for VS15, and after creating a VM with the required dependencies / requirements the build still failed.  (You might have more insight/luck.)
+ I appreciate the effort/time the team put into this, yet I'm giving you a head ups if you investigate their resources further.

Open ALPR requires dependencies not included / built into OpenAlprClient, the files are included on the build:
```
  <ItemGroup>
    <_CopyItems Include="$(SolutionDir)\MlNetOnnxAlpr.OpenAlprClient\Dependencies\**\*.*">
      <InProject>false</InProject>
    </_CopyItems>
  </ItemGroup>
  <Target Name="PostBuild" AfterTargets="PostBuildEvent">
    <Copy SourceFiles="@(_CopyItems)" DestinationFiles="@(_CopyItems->'$(OutDir)\%(RecursiveDir)%(Filename)%(Extension)')" SkipUnchangedFiles="true" OverwriteReadOnlyFiles="true" Retries="3" RetryDelayMilliseconds="300" />
  </Target>
```

##### .NET Core & .NET Framework woes
+ Yes, it's a .NET Core application calling a .NET Framework DLL which is quite bad, yet you could create an API / micro-service to pass the bitmap too and return the relevant information, for a production end-to-end solution.

#### Background

+ Had the idea with a colleague, using car dash cam(s) to determine the vehicle type/information, as well as log the license plate information.
+ Realized it was possible with this research publication [Microsoft CNN-Automatic License Plate](https://www.microsoft.com/en-us/research/publication/a-cnn-based-approach-for-automatic-license-plate-recognition-in-the-wild/)
+ Found a few projects by MicrocontrollersAndMore [OpenCV_3_License_Plate_Recognition](https://github.com/MicrocontrollersAndMore/OpenCV_3_License_Plate_Recognition_Emgu_CV_3_Visual_Basic) (the source is very rough yet functional)

#### Completed & Better Solutions / Alternatives

+ [Deep ANPR - Python](https://github.com/matthewearl/deep-anpr)
+ [Alpr-Unconstrained - Python](https://github.com/sergiomsilva/alpr-unconstrained)


#### Resources

+ [Netron - Onnx Model Viewer](https://github.com/lutzroeder/netron)