31-08-2017, 10:55 AM
Mobile augmented reality (MAR) is a broad class of applications where mobile devices increase users' perception of the world. Many mobile phones that capture video images or still images from a scene can automatically recognize and annotate objects in the scene. Existing MAR systems include benchmark recognition, product logo recognition, and CD / DVD cover recognition. The real-time increase in a telephone is still difficult because a MAR system has delays in
(1) retrieving query data on the phone,
(2) transmitting the query data from the phone through a wireless network to a server that hosts an image database, and
(3) search the database.
Real-time recognition requires small delays in all three stages, while ensuring high recognition accuracy. MAR systems presented in all require delays of at least 3 seconds to recognize an object that appears recently. For the continuous increase of live video, the recognition latency should be reduced to approximately 1 second. The system achieves a delay of approximately 1 second, but at the expense of the continuous transmission of a video from the mobile phone to the server. In this work, a new MAR system is presented for the continuous recognition of book covers and live video CD captured by a mobile phone, an innovation we refer to as MAR streaming. The user can point the camera to a book or CD and view the identity in the viewfinder in about 1 second. The object boundary is displayed and accurate tracking is performed in real time. Both the identity and geometry of the object are quickly retrieved from a server that hosts a database of 20,000 images of books and CDs. As the user advances through the scene, the system automatically recognizes new objects that enter the view, without the user having to press a button. Unlike our system performs motion analysis on the phone and selectively decides when to send new query data, instead of continuously transmitting video over a wireless network.
(1) retrieving query data on the phone,
(2) transmitting the query data from the phone through a wireless network to a server that hosts an image database, and
(3) search the database.
Real-time recognition requires small delays in all three stages, while ensuring high recognition accuracy. MAR systems presented in all require delays of at least 3 seconds to recognize an object that appears recently. For the continuous increase of live video, the recognition latency should be reduced to approximately 1 second. The system achieves a delay of approximately 1 second, but at the expense of the continuous transmission of a video from the mobile phone to the server. In this work, a new MAR system is presented for the continuous recognition of book covers and live video CD captured by a mobile phone, an innovation we refer to as MAR streaming. The user can point the camera to a book or CD and view the identity in the viewfinder in about 1 second. The object boundary is displayed and accurate tracking is performed in real time. Both the identity and geometry of the object are quickly retrieved from a server that hosts a database of 20,000 images of books and CDs. As the user advances through the scene, the system automatically recognizes new objects that enter the view, without the user having to press a button. Unlike our system performs motion analysis on the phone and selectively decides when to send new query data, instead of continuously transmitting video over a wireless network.