Real-time logo detection and tracking on PC and open source mobile platforms

This thesis involves the real-time implementation of a logo detection and tracking algorithm on two platforms. The algorithm uses a hybrid approach by running the Scale Invariant Feature Transform (SIFT) on the first frame to obtain the location of a logo and then uses an online calibration of color within the SIFT detected area to track the logo in subsequent frames in a time efficient manner. The first part of the thesis addresses the real-time implementation of this hybrid approach on the PC platform. The algorithm is then ported onto typical open source mobile platform, namely the BeagleBoard. The processing power of this open source platform is typical of that of handheld devices. Various optimization steps taken to allow the real-time execution of the algorithm on the mobile platform are discussed. The results obtained indicate that real-time robust logo detection and tracking can be achieved on such limited resources platforms.