See the GitHub repository here.

For my university third year dissertation project, I developed a computer vision-orientated application for detecting and classifying bottles from a camera stream or image frame, as well as displaying appropriate cocktail recipes based on the classified bottles in an aesthetically pleasing web page.

This was achieved using trained HAAR/LBP cascades (using the OpenCV implementations) for detection and colour histograms generated from detected regions of interest to improve classification performance. The applications were written primarily in Python utilising a wide number of libraries (list can be seen on GitHub repository), though web languages were also used for rendering the resolved recipes. The tool currently classifies Smirnoff Vodka and Jack Daniel’s bottles, as well as Red Bull cans, Monster Energy cans and an orange juice carton. MySQL was used to store the scraped cocktail recipes.

Several command line tools were developed which allow the flexible execution and maintenance of the tool:

• Recipe Scraper: used to scrape a recipe by name or multiple recipes by ingredient name from a cocktail recipe website into the database.
• Histogram Generator: used to generate specially formatted histogram files which represent the colour spectrum of the target image.
• Main Application: used to perform classification of the target image frame or camera stream. Also used to resolve the cocktail recipes and to launch the web server for hosting the recipe view render.

Screenshots of the classification of bottles from a live webcam feed:

Screenshots of the cocktail recipe lists resolved from classified bottles: