Researchers from the Jožef Stefan Institute’s Smart Systems Department won the open world championship on recognizing activity using sensors in a smartphone. The competition was organized by the University of Sussex (UK) in collaboration with the Chinese company Huawei. The championship winners were named during the Ubicomp 2018 Conference held in Singapore.
In total, 35 teams from all over the world participated in the championship called Sussex-Huawei Locomotion and Transportation Challenge, but only 19 of them were able to cope with the difficult task. Only two methods suggested by the Slovenian researchers allowed over 90% of correct answers.
The competition task was based on video signals from mobile phones of three persons recorded over seven months. Every day they manually recorded the way they travelled, namely: standing, walking, running, cycling, driving, using a bus, a train and subway.
Based on the video, the participants of the championship tried to develop a method of recognizing the mode of locomotion and transportation activities. At the next stage, they had to independently recognize the means of movement on new, not yet marked data using the indicated method. The marking parameters were known only to the organisers, who compared the participants’ forecasts with actual data and thus determined the winners.
Solutions suggested by the researchers from the Jožef Stefan Institute’s Smart Systems Department are based on the principle of multiple knowledge, many years of experience and a wide search for optimal parameters. For each final forecast several separate local forecasts were made. Then all the forecasts were combined into one general forecast. When making forecasts, deep neural networks, decision trees, forest of trees, stimulation, and other methods were used.
The organisers of the competition opened access to the five best methods. This means that results will be used not only by Huawei. Experience shows that the best approaches quickly come into general use on the majority of smartphones.