Impaired gait balance control is associated with an increased risk of falls and reduced mobility, particularly in older adults. Traditional methods of assessing gait balance control, such as clinical balance assessments and biomechanical analysis, have limitations in terms of reliability, cost, and practicality. Wearable sensor technology, including IMUs, offers a more accessible and cost-effective alternative for assessing gait and balance performance in real-world settings. In this investigation, we first examined the similarity between the COM accelerations calculated by conventional motion capture systems and the accelerations directly measured with the accelerometer placed at L5 during level walking. The study further analyzed the IMU signals obtained from different gait activities, obstacle crossing and stair ascending/descending, to characterize balance perturbations. We also employ machine learning algorithms to distinguish different gait activities and detect gait imbalance.