MCRC-SCUT合作小组开发实时行为检测演示系统

 

A new kind of interaction technology that understands users’ movement has emerged in recent years due to the rapid development of accelerometer sensor technology. A joint research team from Motorola China Research Center (MCRC) and South China University of Technology (SCUT) has recently developed a real time human activity recognition system using ADI tri-axial accelerometer chip. This system has a wireless implementation that sends data from acceleration sensing unit via Bluetooth to a Pocket PC. An activity-recognition software developed on Pocket PC by MCRC-SCUT team automatically extracts the data, analyzes them and then display the final recognized results in real time.

   Key algorithms of the activity recognition system include data pre-processing, feature analysis and extraction, and fast decision tree classification. One innovation contribution of the system is that a new weightlessness-based feature is extracted for the analysis and recognition of human activity, based on the observation that there exists instantaneous weightlessness phenomenon in many human activities such as running and jumping. Using only a few dimensions of features, the system can robust recognize four kinds of human activities: walking, running, jumping and still, in real time by a fast decision tree classifier with more than 98% accuracy. The system can also be used as a pedometer which is able to count user’s steps, estimate the speed of movement and the energy consumption in calorie. Furthermore, with the new weightlessness feature, the system can even estimate the jumping height.

The system does not require the accelerometer sensor tied to a special body location (the sensor can be just naturally put in user’s pocket of trousers or jacket), and it is not sensitive to small rotation variations of the sensor, therefore making it more appealing and useful when the accelerometer sensor becomes part of the cell phone in future practical applications.