Skip to content

Latest commit

 

History

History
89 lines (69 loc) · 4.24 KB

README.md

File metadata and controls

89 lines (69 loc) · 4.24 KB
Raw

Object Detection with OpenCV and SSD MobileNet v3

Description

Screenshot 2024-06-19 121308 Screenshot 2024-06-19 121106

This Python script leverages OpenCV and a pre-trained SSD MobileNet v3 deep learning model for real-time object detection in video streams. It utilizes the COCO (Common Objects in Context) dataset for object class recognition.

Requirements

  • Python 3.x (Download)
  • OpenCV library: Install via pip install opencv-python
  • NumPy library: Install via pip install numpy (usually included with OpenCV)
  • coco.names file: Contains class labels for the COCO dataset (download from a reliable source)
  • ssd_mobilenet_v3_large_coco_2020_01_14.pbtxt file: Model configuration file (download from a reliable source)
  • frozen_inference_graph.pb file: Model weights file (download from a reliable source)

Instructions

  1. Download necessary files:

    • Obtain the coco.names, ssd_mobilenet_v3_large_coco_2020_01_14.pbtxt, and frozen_inference_graph.pb files from a trusted source (ensure compatibility with your OpenCV version). Place them in the same directory as your Python script.

  2. Run the script:

    • Open a terminal or command prompt, navigate to the directory containing your script and files.
    • Execute the script using: 
      python object_detection.py

Code Breakdown

  1. Imports:

    • cv2: Imports the OpenCV library for computer vision tasks.

  2. Threshold and Video Capture:

    • thres: Sets the confidence threshold for object detection (adjust as needed).
    • cap: Initializes a video capture object (VideoCapture(1)) to access your webcam (or a different video source by providing its index or path).
    • cap.set(): Sets video capture properties:
      • 3: Width (adjust for desired resolution)
      • 4: Height (adjust for desired resolution)
      • 10: Brightness (adjust for lighting conditions)

  3. Load Class Names:

    • classNames: Creates an empty list to store object class names.
    • classFile: Path to the coco.names file.
    • Loads class names from the file and splits them into a list.

  4. Load Model Configuration and Weights:

    • configPath: Path to the ssd_mobilenet_v3_large_coco_2020_01_14.pbtxt file.
    • weightsPath: Path to the frozen_inference_graph.pb file.
    • net: Creates a detection model object using cv2.dnn_DetectionModel().
    • Sets model input size, scale, mean, and color channel swapping parameters for compatibility with the model.

  5. Main Loop:

    • while True: Continuously captures frames from the video stream.
    • success, img: Reads a frame and checks for success. Exits if unsuccessful.
    • classIds, confs, bbox: Performs object detection using the model on the current frame.
      • classIds: List of detected object class IDs.
      • confs: List of corresponding confidence scores (0-1).
      • bbox: List of bounding boxes (coordinates) for detected objects.
    • Prints detected object IDs and bounding boxes for debugging (optional).

  6. Draw Bounding Boxes and Labels:

    • Checks if any objects were detected (len(classIds) != 0).
    • Iterates through detected objects using zip:
      • box: Current bounding box coordinates.
      • classId: Current object class ID (minus 1 for indexing).
      • confidence: Current object confidence score.
    • Draws a green rectangle around the detected object using cv2.rectangle().
    • Displays the corresponding class name (uppercase) and confidence score (rounded to two decimal places) using cv2.putText().

  7. Display and Exit:

    • cv2.imshow(): Displays the processed frame with bounding boxes and labels in a window titled "Output".
    • cv2.waitKey(1): Waits for a key press for 1 millisecond.
    • Exits the loop and releases resources if the 'q' key is pressed.

  8. Cleanup:

    • Releases the video capture object and closes all OpenCV windows. 
      cap.release()
cv2.destroyAllWindows()

License

This project is licensed under the MIT License. thannk