We present and evaluate a method to generate a compressed representation of multi-dimensional inertial sensor signals using a piecewise linear approximation. The representation can be computed on small sensor nodes and thus allows for a reduction of the amount of data that needs to be transmitted to the main processing node. On an existing gesture database, we present the compression rate that is reached and evaluate the quality of the representation in terms of the accuracy reached for gesture classification. We compare the results to our baseline system using a simpler approach for data reduction.