Real-time Colour Recognition

Overview

A computer vision application that detects, marks and classifies colours in camera video in real time. Built with Python and OpenCV.

The Need

Colour-based object detection is a building block for many applications, including quality control, sorting and interactive systems. Colour perception, however, is highly sensitive to lighting: the same object produces entirely different pixel values under different illumination. The goal was a system that delivers consistent results under changing conditions and can be calibrated quickly on site.

The Solution

Every frame passes through a five-stage image processing pipeline:

1. Pre-processing — Gaussian blur reduces sensor noise.
2. Colour space conversion — The image is converted from RGB to HSV. Separating hue from brightness provides marked robustness to lighting changes.
3. Masking — Binary masks are created from defined HSV ranges for each target colour; for colours such as red, which wrap around both ends of the hue circle, ranges are combined.
4. Morphological refinement and contour analysis — Opening and closing operations remove noise, contours are extracted and regions below a size threshold are discarded.
5. Visualisation — Each detected object is framed with its colour label, and the dominant colour at the centre of the image is reported separately.

On-site Calibration

Anticipating that fixed thresholds would not hold in every environment, a calibration interface was built to adjust HSV bounds live on the camera image. The system can be reconfigured for a new environment within seconds.

Impact

• A lightweight, explainable detection pipeline that is robust to lighting changes.
• Processing steps optimised with frame rate in mind for real-time performance.
• A solid foundation for subsequent machine learning and computer vision work.