Real-Time Object Detection with EfficientDet Lite and OpenCV

This project demonstrates real-time object detection using a webcam feed, leveraging OpenCV and MediaPipe. The EfficientDet Lite0 model is used to detect objects, and bounding boxes are drawn around them, with class labels displayed on each frame.

Requirements

Before running the project, ensure that the following Python libraries are installed:

• opencv-python: For video capture and image processing.
• mediapipe: For vision tasks, including object detection.

To install the dependencies, run:

pip install mediapipe opencv-python

Alternatively, you can install all dependencies listed in the requirements.txt file:

pip install -r requirements.txt

Ensure the efficientdet_lite0.tflite model is placed in a models directory, as specified in the code, for object detection functionality.

Functions

getObjects(frame, textColor=(0, 255, 0))

This function processes an input frame and detects objects, drawing bounding boxes around each detected object.

Parameters:

• frame (numpy.ndarray): The image frame from the webcam feed, video, or image.
• textColor (tuple, optional): The color of the text (default is green).

Returns:

• A frame with bounding boxes drawn around detected objects.

main()

This function continuously captures video from the webcam or loads an image/video file, applies object detection on each frame, and displays the result. The stream will stop when the user presses the 'q' key or reaches the end of the video.

How It Works

1. Object Detection: The EfficientDet Lite0 model, loaded using MediaPipe's ObjectDetector, processes each frame to identify objects in real time.
2. Bounding Boxes: For each detected object, a bounding box is drawn using OpenCV’s cv2.rectangle(). The object's class label is displayed near the bounding box.
3. Displaying Results: The processed frame, now containing the bounding boxes and labels, is continuously displayed in a window.

Running the Program

The program accepts command-line arguments to specify the input type (image, video, or webcam), input file, and output options. Below are the details for using these arguments:

Arguments

• --input-type / (-t): Required
  Specifies the type of input:

  • image: Use an image file as input.
  • video: Use a video file as input.
  • webcam: Use a webcam for live video input.

• --index / (-i):
  Specifies the webcam index (only for webcam input). If you have multiple webcams, you can specify which one to use (default is 0).

• --input-file / (-if):
  Specifies the path to the input image or video file. This is required when input-type is either image or video.

• --output-file / (-of):
  Specifies the path of the output file where the result will be saved. This can be an image file (for image input) or a video file (for video or webcam input). If not provided, the processed frames will only be displayed.

Examples

For Webcam Input:

To start real-time object detection using the webcam, run:

python object_detection.py --input-type webcam

This will use the default webcam (index 0).

You can use the --index / -i argument if you want to specify a specific webcam when the input type is set to webcam. For example:

python object_detection.py --input-type webcam --index 1

This will use the second webcam (if available) for object detection.

For Image Input:

To run object detection on an image, provide the path to the image file using the --input-file / -if argument:

python object_detection.py --input-type image --input-file path_to_image.jpg --output-file output_image.jpg

This will detect objects in the specified image and save the processed image to output_image.jpg.

For Video Input:

To run object detection on a video file, provide the video file path using the --input-file / -if argument:

python object_detection.py --input-type video --input-file path_to_video.mp4 --output-file output_video.mp4

This will process the video and save the output with the detected objects to output_video.mp4.

Note-

• If you choose to use a video or image file as input, make sure to specify the path to the input file using the --input-file / -if argument.

• The output file is optional. The processed image or video or webcam feed will be displayed live unless you specify an output file to save the result using --output-file / -of.

• The program uses efficientdet_lite0.tflite, which should be placed in a models directory to function correctly.

• Press 'q' to exit the program.