Location-aware mobile information hub

• Main Author: Xie, Xiaohui.
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/53058
• Institution: Nanyang Technological University

With the rapid development of mobile devices and image processing technologies, mobile visual search has become a hot topic in both research and commercial fields. Nonetheless, some of its application domains such as landmark visual search still pose challenges in terms of recognition accuracy and speed. For a typical landmark recognition system that is based on Vocabulary Tree (VT), Geometric Verification (GV) is often added as the final stage to boost the usually low recognition accuracy. However, GV is very computationally expensive and thus it significantly increases the recognition latency. In view of this, the author’s project incorporates an algorithm called Fast Geometric Re-Ranking (FGRR) into the current recognition system, with the objectives of speeding up the system while maintaining or even improving the accuracy performance. Experimental results have shown that FGRR is able to fulfill such requirements. In terms of accuracy performance, FGRR is able to improve the overall recognition rate by around 1% to 5%, depending on whether our system employs GV. In terms of speed performance, FGRR takes only a small amount of time in the order of 0.001 seconds per query image, which is only a tiny portion compared to the total recognition latency. It is also shown that for our system employing GV, FGRR is able to save up to 2 seconds or more per query image and meanwhile maintain the same level of accuracy performance.