Real-Time Product Detection with Annotation using Features Detection and Tracking on Graphics Processing Unit (GPU)
This project is part of the bigger Hi-PoP project under the Institute for Media Innovation (IMI) grant. We combine the novel use of feature tracking and augmented reality into useful commercial application – in this case it provides shopper with new interactive media experiences.
The system provides shoppers with production information at their fingertips. When viewed through a camera, shoppers are able to see virtual tag of the product overlaid on the product itself, in an augmented reality environment. In a nutshell, the camera tracks useful features on the product label and the system renders 3D virtual information onto it, giving shoppers a 3D mixed reality view of the real/physical product and virtual information.
The consumer’s preferences of products are profiled and relevant product recommendation can be made. The use of this are, but not limited to: virtual price tag, and product branding and promotion.
Approach
1.Use features detection and extraction to identify product.
2.The features detecting and matching is done on GPU for high performance and speed.
. 
Why GPU-based Feature Tracking?
1.Arrival of dedicated graphics hardware with increasing computational power and programmability.
2.Suitable for parallel processing of extensive calculations.
3.Extracted Feature Points are scale and rotation invariant.
4.Also robust to changes in illumination, noise, and minor changes in viewpoint.
Feature Detection and Matching
1.Scale Invariant Feature Transform (SIFT) algorithms is used
2.Features points are extracted from the input images.
3.Match feature points from camera image with the feature points of input images.
4.Detect and find the input image with the most matches
Results
Tracking and annotation of the product.
. 
. 
. 
Conclusion
1.Problem with some products as the features detected are insufficient for matching.
2.The material of product packaging is too glossy and causes reflection. Hence the image is blurred.
3.Problem with jittering of the virtual object.
4.Increase the frame rate by improving the speed of detection and matching process.
