A novel one-pass neural network approach for activities recognition in intelligent environments

Designing less intrusive intelligent environments requires a deep understanding of activities that a user is engaged in. This paper presents a novel one-pass neural network system that uses unobtrusive and relatively simple sensors and puts forward a constructive algorithm which is able to recognize different high level activities (such as ldquosleepingrdquo, ldquowashingrdquo, ldquoworking at computerrdquo) in intelligent inhabited environments. The neural network system adding temporal capabilities is able to recognize abnormal behaviors. One-pass learning method of weight ratios can rapidly improve the learning speed and reduce the memory of embedded computer. It can be trained in an online mode and hence it can be integrated into the limited processor-power embedded computing platforms used in intelligent environments. Experiment results show that this method is transparent, simple and effective.