Kinematic modelling of a walking cycle in an unmonitored environment
This report reviews the progress of work done on our bachelor's degree final project. The goal of this project is to design a system of kinematic modelling of a walking cycle in an unmonitored environment. Several challenges needed to be overcome in this project and the considerations and solutions found during the development process is discussed in this report.
The suggested system is a sensor suit and its development demanded both hardware and software characterization and fabrication. The sensing part of the system is achieved using seven IMU's (Inertial Measurement Units), which are capable of measuring angular velocity, linear acceleration and the magnetic field in three orthogonal axis.
During this project we understood the pros and cons of using each one of the measured physical data records which are required in order to produce the orientation of each measuring unit. After considering several methods that could have been used to produce the orientation of the measuring unit, without drifts over time and with acceptable accuracy, we came to the conclusion that the best option is the Mahoney Filter. In order to communicate between the many units in the system, three methods of communication are integrated into the system: SPI, I2C and Bluetooth, the former done via Xbee module. At the moment this is a bottle neck of the system's sampling frequency. Once we completed the hardware development phase and we were able to obtain raw data from the system, we focused on writing algorithms that provide the needed data that will allow gait cycle analysis.
There are two major phases in the gait cycle:
Heel strike- when the heel hits the ground and Toe off- when the toes leaves the ground. Due to relative high forces and moment during these phases, recognizing the loads which apply while they occur is crucial for recognition of causes lower body injuries. An algorithm we developed uses the measured acceleration and the produced orientation of each part of the lower body to recognize the mentioned phases, in real time. In order to produce the 3D kinematic model of the lower body while walking, we developed an algorithm which refers to the human's lower body as an open kinematic chain. By knowing the orientation of each of the chain’s links, the position and orientation of the entire chain is produced.
To validate the orientation we obtained from each measuring unit, we compared the data that was obtained using to the QTM (Qualisys Tracking Manager). This is considered as a "golden standard" of the field of body motion capture. In the test we found that the RMSE between the two systems is 1.3 degrees, which is highly acceptable. The results of this test also thought us that our system is sensitive to local magnetic fields and to accelerations during Heel strike. To deal with errors during heel strike we decided to use a dynamic proportional gain constant (Kp). The value of Kp is reduced during heel strike so while calculating the orientation with the Mahoney filter, more lower weight is given to the acceleration than to the angular velocity compared to other stages of the gait cycle. In the future, another algorithm will be added to the system to calculate the length and width of each step, based on the position that is obtained by the kinematic chain.