On-line measurement of respiration plays an important role in monitoring human physical activities. Such measurement commonly employs sensing belts secured around the abdomen of the test object. This paper first presents a signal decomposition technique for tissue artifact removal from respiratory signals and respiratory signal reconstruction, based on the Empirical Mode Decomposition (EMD). Methods based on spectral analysis and multiple linear regressions were then developed to predict the respiration rate and minute ventilation, respectively. Performance of the algorithms was evaluated through real-life experiments of 105 subjects engaged in 14 types of physical activities. The predictions were compared to the criterion respiration measurements using a bidirectional digital volume transducer housed in a respiratory gas exchange system. Results have verified reasonably good performance of the algorithms and the applicability of the wearable sensing system for respiratory parameter prediction during physical activity.