Statistical data mining of streaming motion data for fall detection in assistive environments

The analysis of human motion data is interesting for the purpose of activity recognition or emergency event detection, especially in the case of elderly or disabled people living independently in their homes. Several techniques have been proposed for identifying such distress situations using either motion, audio or video sensors on the monitored subject (wearable sensors) or the surrounding environment. The output of such sensors is data streams that require real time recognition, especially in emergency situations, thus traditional classification approaches may not be applicable for immediate alarm triggering or fall prevention. This paper presents a statistical mining methodology that may be used for the specific problem of real time fall detection. Visual data captured from the user's environment, using overhead cameras along with motion data are collected from accelerometers on the subject's body and are fed to the fall detection system. The paper includes the details of the stream data mining methodology incorporated in the system along with an initial evaluation of the achieved accuracy in detecting falls.

EFFICIENTLY MATCHING FREQUENT PATTERNS BASED ON BITMAP INVERTED FILES BUILT FROM CLOSED ITEMSETS

Online pattern matching has become an increasingly important method for utilizing the discovered frequent patterns to build various intelligent systems, and obviously it demands high pattern matching efficiency. Memory-based online pattern matching is a solution, but it requires a frequent pattern base with small storage size. The closed itemsets are a lossless and condensed representation of all frequent itemsets, however, their itemset format does not facilitate implementing efficient pattern matching. This paper proposes a novel method that builds the bitmap inverted file from the closed itemsets and uses it to substitute for the closed itemsets to perform pattern matching. The bitmap inverted files use bitmaps instead of the referential lists to keep the maps of frequent items in the closed itemsets, so as to reduce the storage size and promote the intersecting operation efficiency. In order to completely substitute for the closed itemsets in pattern matching, the bitmap inverted files also keep other information relevant to the closed itemsets, such as their supports and lengths. Experiments show that on dense datasets, the storage size of the bitmap inverted files is much smaller than that of the set of closed itemsets, and the pattern matching based on the bitmap inverted files resident in memory is orders of magnitude more efficient than that based on the closed itemsets.