grape-detect

Model weights can be found here: https://drive.google.com/file/d/1wz0pACVMTV9x1HwatQSxKj8wSWG5yOcQ/view?usp=sharing

To perform object detection, run the grapes_yolov3.py file.

Description

This code defines a Python class called StreamCamera that performs real-time object detection on a video stream from a camera. The class uses the YOLOv3 (You Only Look Once) deep learning model to detect grapes in the video frames. It keeps track of the number of grapes detected and calculates a yield estimate based on the count of detected grapes.

Here is a summary of the code:

• The code imports the necessary libraries: cv2 for computer vision operations and numpy for numerical computations.
• The StreamCamera class is defined, which has the following attributes:
  • camera_index: An index representing the camera source to capture the video stream.
  • running: A boolean variable indicating whether the video stream is running.
  • net: The YOLOv3 neural network model.
  • classes: A list of classes that the model can detect.
  • detected_grapes: A set to store the unique identifiers of detected grapes.
• The StreamCamera class provides the following methods:
  • load_model(): Loads the YOLOv3 network from configuration and weight files and reads the classes from a file.
  • start_stream(): Starts the video stream, continuously reads frames from the camera, detects grapes in each frame, and displays the frames with bounding boxes and labels. It also calculates the yield estimate.
  • detect_objects(frame): Performs grape detection on a given frame using YOLOv3. It processes the frame through the network and extracts the bounding boxes, class labels, and confidence scores of the detected grapes.
  • draw_detections(frame, detections): Draws bounding boxes, labels, and circles around the detected grapes on the input frame. It also adds the unique identifier of each detected grape to the set of detected_grapes.
  • calculate_yield_estimate(): Calculates the yield estimate by counting the number of unique detected grapes.
  • stop_stream(): Stops the video stream by setting the running attribute to False.
  • The main() function creates an instance of StreamCamera with a camera index of 0 (typically representing the default camera) and starts the video stream by calling the start_stream() method.
• The script is executed if it is run directly (i.e., not imported as a module) by calling the main() function.

In summary, this code sets up a real-time grape detection system using YOLOv3 and estimates the yield of grapes based on the detected count in the video stream.