Logo Recognition Using Deep Learning and Storing Screen Time in MongoDB Database

A Logo is the medium through which a company reaches its target audience. Huge amount of funds is invested by companies for every second of screen time of their logo. This project proposes to detect three logos: Adidas, Coca Cola and DHL from various media covered events such as sports or entertainment and store their screen time. A custom dataset containing 1025 images was constructed from sports and entertainment videos. Each image was annotated to its specific logo class. Image preprocessing is done before the images are fed into the Deep Learning model. The model uses Single Shot Multibox Detector (SSD) algorithm along with Inception v2 as a base network to construct the model via Transfer Learning. After training the model, it is used to detect logos in sports videos and the detected logos are stored in a MongoDB database along with their screen time and frequency of occurrence.