基于YOLOv5人脸和CRNN模型的自动车牌识别系统

# **Automatic Number Plate Detection system** github:[https://github.com/201918010332Thomas/ANPR](https://github.com/201918010332Thomas/ANPR) **Software environment requirements: python >=3.6 pytorch >=1.7** ## **GUI program:** ``` python anpr.py ``` ## License plate detection training 1. **Dataset** This project uses open source datasets CCPD and CRPD. The dataset label format is YOLO format： ``` label x y w h pt1x pt1y pt2x pt2y pt3x pt3y pt4x pt4y ``` The key points are in order (top left, top right, bottom right, bottom left). The coordinates are all normalized, where x and y are the center point divided by the width and height of the image, w and h are the width and height of the box divided by the width and height of the image, ptx and pty are the key point coordinates divided by the width and height. 2. **Modify the data/widerface.yaml file** ``` train: /your/train/path #This is the training dataset, modify to your path. val: /your/val/path #This is the evaluation dataset, modify to your path. # number of classes nc: 2 #Here we use 2 categories, 0 single layer license plate 1 double layer license plate. # class names names: [ 'single','double'] ``` 3. **Train** ``` python train.py --data data/widerface.yaml --cfg models/yolov5n-0.5.yaml --weights weights/plate_detect.pt --epoch 250 ``` The result exists in the run folder. 4. **Detection model onnx export** To export the detection model to onnx, onnx sim needs to be installed. **[onnx-simplifier](https://github.com/daquexian/onnx-simplifier)** ``` 1. python export.py --weights ./weights/plate_detect.pt --img 640 --batch 1 2. onnxsim weights/plate_detect.onnx weights/plate_detect.onnx ``` **Using trained models for detection** ``` python detect_demo.py --detect_model weights/plate_detect.pt ``` ## License plate recognition training The training link for license plate recognition is as follows: [License plate recognition training](https://github.com/201918010332Thomas/CRNN_LPR) #### **The results of license plate recognition are as follows:**  ## Arrange 1.**onnx demo** The onnx model can be found in [onnx model](https://pan.baidu.com/s/1UmWN2kpRP96h2cM6Pi-now), with extraction code: ixyr python onnx_infer.py --detect_model weights/plate_detect.onnx --rec_model weights/plate_rec.onnx --image_path imgs --output result_onnx 2.**tensorrt** Deployment can be found in [tensorrt_plate](https://github.com/we0091234/chinese_plate_tensorrt) 3.**openvino demo** Version 2022.2 ``` python openvino_infer.py --detect_model weights/plate_detect.onnx --rec_model weights/plate_rec.onnx --image_path imgs --output result_openvino ``` ## References * [https://github.com/deepcam-cn/yolov5-face](https://github.com/deepcam-cn/yolov5-face) * [https://github.com/meijieru/crnn.pytorch](https://github.com/meijieru/crnn.pytorch)