Research On Mobile Visual Location Recognition Based On Fusion Of Visual And Sensors

With the development of hardware technology, ordinary smart phones have beenable to carry out complex image processing tasks, thus promoting the emergence anddevelopment of mobile visual location recognition technology. Current mobile visuallocation recognition research is based on the client-server mode, using the visualcharacteristics to recognize the image target. This mode is bound to generate networklatency, limited storage capacity of the client, limited compute power of the clientand weak discrimination of the city-level database visual characteristics.For a general lack of geographic information marked in the existing city-levelimage database, the Wuhan urban street view image database is constructed whichcontains1,295,000image data. The geographical coordinates and directioninformation is recorded with each database image which is grouped by GPS. Sincemobile visual location recognition needs to retrieve large volumes of data, the sensorsof a mobile device are used to filter the search range. The selected GPS distance isless than200meters and the direction of the difference within60degrees of thedatabase image retrieval, which excludes a large number of non-related images, notonly reducing the amount of retrieved data, but also improving the retrieval accuracy.To solve the problem of computational complexity for the visual features in imageretrieval stage, the speeded upright SURF features are extracted to generate two kindsof image descriptors, the binary codes are computed respectively, the multi featuresare fused to improve the resolution of the image retrieval,128dimensions of theimage descriptor fused are selected to describe the visual features of each GPSpartition, and GPS, direction and selection of visual features are fused to generate theindex files.Experimental results show that the introduction of the sensor can reduce theamount of retrieving data and reduce the retrieval time. The computed binary codes,the integration of visual characteristics and the selection dimension of the imagedescriptor can compress images, accelerate image similarity calculation, and ensurethe correct rate of the mobile visual location recognition results at the same time. The index files established based on fusion of sensors and vision can save the phone’sstorage space and computing time, which completes the visual location recognitiondirectly on mobile devices.