BSpace: A Data Cleaning Approach For RFID Data Streams Based On Virtual Spatial Granularity

In recent years, with the development of micro-electronics, computer and network technologies, RFID technology has been applied widely and deeply. RFID (Radio Frequency Identification) is a non-contact automatic identification technology, which aims at identifying and tracking items by using radio frequency electromagnetic wave to let readers capture the data on RF tags attached to them. However, unreliable data (original data) captured by readers is a major factor hindering the development of RFID technology. Unreliability of RFID data is classified into three categories: false positives, false negatives, and duplicates. For effectively and efficiently supporting high-level RFID business logic processing, it is necessary to provide high-quality RFID data. For that case, it is critical to clean the original data.In this paper, according to our in-depth research on RFID data stream cleaning, we propose a new cleaning approach based on the virtual spatial granularity, named bSpace, specific working is as follows:Aiming at coping with weaknesses of the sliding window and the traditional spatial granularity, we firstly propose the concept of virtual spatial granularity and the method that we use to obtain the virtual spatial granularity. During the processing of cleaning, the size of the virtual spatial granularity can be adjusted automatically and continuously based on the real-time observation on RFID data.In order to solve false positives, this paper uses the rules which we define based on virtual spatial granularity; In order to fill up false negatives for dynamic tags, for a tag present in the detection range of a reader, bSpace uses a Bayesian estimation strategy to compute the times that the tag has been detected; Theoretically, the detected data from tags in the same virtual spatial granularity are as same as the real data, so operation OR on these tags can realize supplement for missing readings to achieve data cleaning. For solving duplicates, we propose the allocation strategy "first the middle, then the two sides", it reallocates the number that we have had in virtue of the estimation , so we can get the time when a tag comes into the detection of a reader and when it leaves, the information between this time will be regard as duplicates.In the end, the experiments prove the reasonable and effectiveness of bSpace, which has realized the cleaning for RFID data efficiently.