Skip to content

The project can achieve FCWS, LDWS, and LKAS functions solely using only visual sensors. using YOLOv5 / YOLOv5-lite / YOLOv6 / YOLOv7 / YOLOv8 / YOLOv9 / EfficientDet and Ultra-Fast-Lane-Detection-v2 .

License

Notifications You must be signed in to change notification settings

jason-li-831202/Vehicle-CV-ADAS

Repository files navigation

Vehicle-CV-ADAS

Python OnnxRuntime TensorRT Markdown Visual Studio Code Windows

Example scripts for the detection of lanes using the ultra fast lane detection v2 model in ONNX/TensorRT.

Example scripts for the detection of objects using the YOLOv5/YOLOv5-lite/YOLOv6/YOLOv7/YOLOv8/YOLOv9/YOLOv10/EfficientDet model in ONNX/TensorRT.

Add ByteTrack to determine the driving direction of ID vehicles and perform trajectory tracking.

➤ Contents

  1. Requirements

  2. Examples

  3. Demo

  4. License

!ADAS on video

➤ Requirements

  • Python 3.7+

  • OpenCV, Scikit-learn, onnxruntime, pycuda and pytorch.

  • Install :

    The requirements.txt file should list all Python libraries that your notebooks depend on, and they will be installed using:

    pip install -r requirements.txt
    

➤ Examples

  • Download YOLO Series Onnx model :

    Use the Google Colab notebook to convert

    Model release version Link
    YOLOv5 v6.2 Open In Colab
    YOLOv6/Lite 0.4.0 Open In Colab
    YOLOv7 v0.1 Open In Colab
    YOLOv8 8.1.27 Open In Colab
    YOLOv9 v0.1 Open In Colab
    YOLOv10 8.2.41 Open In Colab
  • Convert Onnx to TenserRT model :

    Need to modify onnx_model_path and trt_model_path before converting.

    python convertOnnxToTensorRT.py -i <path-of-your-onnx-model>  -o <path-of-your-trt-model>
    
  • Quantize ONNX models :

    Converting a model to use float16 instead of float32 can decrease the model size.

    python onnxQuantization.py -i <path-of-your-onnx-model>
    
  • Video Inference :

    • Setting Config :

      Note : can support onnx/tensorRT format model. But it needs to match the same model type.

    lane_config = {
     "model_path": "./TrafficLaneDetector/models/culane_res18.trt",
     "model_type" : LaneModelType.UFLDV2_CULANE
    }
    
    object_config = {
     "model_path": './ObjectDetector/models/yolov8l-coco.trt',
     "model_type" : ObjectModelType.YOLOV8,
     "classes_path" : './ObjectDetector/models/coco_label.txt',
     "box_score" : 0.4,
     "box_nms_iou" : 0.45
    }
    Target Model Type Describe
    Lanes LaneModelType.UFLD_TUSIMPLE Support Tusimple data with ResNet18 backbone.
    Lanes LaneModelType.UFLD_CULANE Support CULane data with ResNet18 backbone.
    Lanes LaneModelType.UFLDV2_TUSIMPLE Support Tusimple data with ResNet18/34 backbone.
    Lanes LaneModelType.UFLDV2_CULANE Support CULane data with ResNet18/34 backbone.
    Object ObjectModelType.YOLOV5 Support yolov5n/s/m/l/x model.
    Object ObjectModelType.YOLOV5_LITE Support yolov5lite-e/s/c/g model.
    Object ObjectModelType.YOLOV6 Support yolov6n/s/m/l, yolov6lite-s/m/l model.
    Object ObjectModelType.YOLOV7 Support yolov7 tiny/x/w/e/d model.
    Object ObjectModelType.YOLOV8 Support yolov8n/s/m/l/x model.
    Object ObjectModelType.YOLOV9 Support yolov9t/s/m/c/e model.
    Object ObjectModelType.YOLOV10 Support yolov10n/s/m/b/l/x model.
    Object ObjectModelType.EfficientDet Support efficientDet b0/b1/b2/b3 model.
    • Run :
    python demo.py
    

➤ Demo

  • Demo Youtube Video

  • Display

    !ADAS on video

  • Front Collision Warning System (FCWS)

    !FCWS

  • Lane Departure Warning System (LDWS)

    !LDWS

  • Lane Keeping Assist System (LKAS)

    !LKAS

➤ License

WiFi Analyzer is licensed under the GNU General Public License v3.0 (GPLv3).

GPLv3 License key requirements :

  • Disclose Source
  • License and Copyright Notice
  • Same License
  • State Changes

About

The project can achieve FCWS, LDWS, and LKAS functions solely using only visual sensors. using YOLOv5 / YOLOv5-lite / YOLOv6 / YOLOv7 / YOLOv8 / YOLOv9 / EfficientDet and Ultra-Fast-Lane-Detection-v2 .

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages