The thesis addresses implementing the SIFT algorithm on an Android platform to detect objects in live video based on reference images stored in memory. The implementation was done in several stages of testing to minimize the number of errors and to make troubleshooting easier and faster. The first step was to assure that the program will work on images alone on a desktop device. The same program was then tested on an Android device and afterwards was upgraded to work on video.
The final program works as intended. However, the object detection is certainly flawed. Not all objects are recognizable and the performance of the algorithm is limited independent of its parameters. Better results would be observed by using better hardware (a better phone). However, the proof of concept is there.
|
