https://github.com/abeed04/object-detection-model-using-cv2

Leverage OpenCV (cv2) to build an object detection system. This system would combine a pre-trained MobileNet model, trained on the COCO dataset, with a frozen inference graph for real-time object identification within images or videos.
https://github.com/abeed04/object-detection-model-using-cv2

cocodataset frozen mobilenetv3 opencv pycharm-ide

Last synced: 15 days ago
JSON representation

Leverage OpenCV (cv2) to build an object detection system. This system would combine a pre-trained MobileNet model, trained on the COCO dataset, with a frozen inference graph for real-time object identification within images or videos.

• Host: GitHub
• URL: https://github.com/abeed04/object-detection-model-using-cv2
• Owner: abeed04
• License: mit
• Created: 2024-06-23T11:16:01.000Z (5 months ago)
• Default Branch: main
• Last Pushed: 2024-07-02T05:31:41.000Z (5 months ago)
• Last Synced: 2024-10-16T22:15:46.001Z (30 days ago)
• Topics: cocodataset, frozen, mobilenetv3, opencv, pycharm-ide
• Language: Python
• Homepage:
• Size: 20 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
Object Detection Model using Cv2 👨‍💻 






 🔭 This project demonstrates real-time object detection using OpenCV (cv2), the MobileNet deep learning model, and the COCO dataset. It leverages a pre-trained frozen inference graph for efficient object identification within images or video streams.

Requirements 

-Python 3.x




-OpenCV (cv2) library




-Pre-trained MobileNet, COCO model files and Frozen model (download them to local pc)




Usage 

-The program automatically attempts to open the video file Streets.mp4(You can upload your own videos). If the file is unavailable, it defaults to your webcam stream.




-A window titled "Object Detection Program" will display the video feed with detected objects highlighted.




-Bounding boxes will be drawn around detected objects, and their corresponding class labels will be displayed.(Note:Only Objects that are labeled in labels.txt file are displayed)




-You can adjust the confidence threshold (confThreshold=0.55) in the code to control the minimum confidence required for a detection to be displayed.




-Press the 'q' key to exit the program.

Explanation 

Imports

-The script imports necessary modules, including OpenCV (cv2)

Model and Label Loading

-The model configuration file (mobilenet_and_coco.txt) and the frozen inference graph (frozen_model.pb) are loaded using cv2.dnn_DetectionModel.




-Class labels are read from labels.txt and stored in a list class_labels.

Model Configuration

-The model's input size is set to 200x200 pixels.




-The input scale and mean values are adjusted for normalization.




-The input color channel order is swapped to BGR (OpenCV's default) if necessary.

Video Capture

-A video capture object cap is created, attempting to open the video file Streets.mp4 first. It then falls back to the webcam if the file is not found.




-Error handling is included to raise an exception if video capture fails.

Main Loop

-The loop continuously reads frames from the video stream using cap.read().




-The frame is resized to match the model's expected input size using OpenCV's resizing functions.




-model.detect is used for object detection in the frame. It returns class indices, confidence scores, and bounding boxes for detected objects that meet the specified confidence threshold.




-For each detected object:





1)A bounding box is drawn around the object using cv2.rectangle.




2)The corresponding class label from class_labels is retrieved and displayed on the frame using cv2.putText.




-The processed frame is displayed in the "Object Detection Program" window.




-The program exits when the 'q' key is pressed.

You can modify the script to:

-The video capture object is released using cap.release().




-OpenCV windows are destroyed using cv2.destroyAllWindows().

Customization 

You can modify the script to:

-Use a different video file or image set for object detection.




-Experiment with different confidence thresholds (confThreshold) to adjust the detection sensitivity.




-Explore ways to integrate this object detection functionality into a larger application.

Further Considerations

-The pre-trained MobileNet model may not be ideal for all object detection tasks. Consider exploring other models or fine-tuning the provided model on your specific dataset for improved accuracy.




-The script currently processes frames one by one. For more efficient real-time performance, you might investigate techniques like multithreading or GPU acceleration.