A detailed review of existing reseach in posture monitoring - PDF
Video - a demonstration of the system is available here.
Human postural and physical activity tracking and monitoring presents many research challenges, partly due to the large number of degrees of freedom of the human body, and partly due to the infancy of the wearable sensing technology for this class of applications. Whilst most wearable networked sensing systems currently in development aim at movement tracking, the work reported here is concerned with the design, deployment and evaluation of postural assessment instrumentation, that can: a) provide the identification and classification of several human postures and b) able to relay this information wirelessly on line and real-time to a remote monitoring point.
Specific applications
During the project lifetime, a variety of applications with varing postural monitoring accuracy needs will be considered. The first of these is the monitoring of personal in Bomb Disposal Missions as part of the Increasing safety in Bomb Disposal Missions project.
Hardware design
The BSN reported here consists of two body mounted nodes (acquisition node [AN] and acquisition & processing node [APN]) and a base station. The Gumstix Verdex XM4-bt devices are used as the main processing and communications platform. The Gumstix devices are fully functional single board computers with a footprint of 80x20x6.3mm^{3} and a weight of 8 grams. The Gumstix devices contain a 400MHz Marvell PXA270 XScale CPU and have integrated Bluetooth communications on-board. This processor board is considerably in excess of the computational requirements for evaluating (not building) a decision tree but the added computational power simplifies the prototyping process, allowing, for example, Python to be used for most of the software development. At the same time, the Gumstix devices are small and light enough to be easily carried in a pouch or pocket.
|
|
compdone
|
|
|
Sensor positions
|
Several volunteers of different builds were used for acquiring training and testing data. The volunteer group was mixed male and female with heights between 1.6m and 1.83m and weights between 60kg and 89kg.
Regime 1: each subject undertook an activity regime composed of sitting, standing, walking, kneeling, crawling, lying on one side, lying down on their front, and lying down on their back. Each posture was maintained for 1 minute, with the subject performing light arm movement tasks combined with variations from the set positions (such as for example, leaning back, forth, sideways, whilst walking and standing).
Regime 2: the performance of the prototype was also evaluated with a mission like activity regime consisting of (1) walking (3 minutes); (2) kneeling while putting weights into and out of a rucksack; (3) crawling (2 minutes); (4) arm exercise while standing (4 minutes); (5) sitting (3 minutes); (6) standing (1 minute). Note that an ideal tree was attempted to be found which, trained on Regime 1 would perform well on mission-like behaviour, such as Regime 2, which is just one example of a mission-like scenario.
Regime 3: this regime expanded on the others described here by adding natural movements to the activities performed (such as lifting weights while standing or packing things into a box while kneeling). Each subject undertook an activity regime composed of standing, running, kneeling, laying on one side, laying down on the front, and laying down on the back. Each posture was maintained for 1 minute.
Performance of was evaluated in terms of postural assessment correctness both for a wide class of wearers with different body builds and for variations in natural movement between subjects. A worst case scenario of 18.5% incorrect posture classifications during a 16 minutes trial was obtained, whilst the mean number of incorrect classifications over 7 such trials was 6.6%.