Object Detection

Category: Mega-Polis → Analysis
Node ID: SvMegapolisObjectDetection
Tooltip: Performs object detection on input images
Dependencies: opencv (cv2), numpy

Functionality

Performs object detection on an input image and outputs detection results such as bounding boxes, class labels, and confidence scores.

The node processes the image using OpenCV-based detection routines (typically pre-trained classifiers or models) and extracts detected object regions for further spatial or analytical workflows.

This node is designed for workflows such as:

  • Detecting objects in street imagery
  • Identifying urban elements (vehicles, pedestrians, façades, etc.)
  • Extracting bounding box geometry for downstream spatial analysis

Inputs

Socket Type Description
Image SvStringsSocket Input image (file path or image array). Must be linked for the node to execute.

Parameters

Name Type Default Description
Confidence Threshold Float 0.5 Minimum confidence score required for detection acceptance.
Draw Boxes Bool True If enabled, outputs image with bounding boxes drawn.

(Exact parameter behaviour depends on the underlying OpenCV detection implementation.)

Outputs

Socket Type Description
Bounding Boxes SvVerticesSocket List of bounding box corner coordinates.
Labels SvStringsSocket Detected object class labels.
Scores SvStringsSocket Confidence scores for each detected object.
Image Out SvStringsSocket Image with detection overlays (if enabled).

Example

Basic object detection workflow

  1. Load an image using an upstream node (e.g., street imagery download).
  2. Connect the image to Image.
  3. Set Confidence Threshold (e.g., 0.6).
  4. Use outputs:
    • Bounding Boxes to generate geometry in the scene
    • Labels to filter specific object types
    • Image Out to visualise detection results

Urban analysis example

  1. Download street imagery.
  2. Run Object Detection.
  3. Count occurrences of certain object classes (e.g., vehicles).
  4. Use counts to drive data visualisation or environmental indicators.

Notes / gotchas

  • Detection quality depends on the pre-trained model used internally.
  • Large images may reduce performance.
  • Bounding box coordinates are typically 2D image coordinates, not world coordinates. ```