A centroid dynamic programming track-before-detect algorithm is proposed, which is applied for tracking the moving target with an unknown speed. Using the inertialess scanning based on an optical phased array, the experimental tracking system is established, and the obtained maximum signal-to-noise ratio is 9.97 dB. Targets of different motion states can be accurately tracked with this algorithm. In addition, we innovated the original track-before-detect algorithm by adding the variable step, so that the target with large accelerations can be tracked accurately. The accuracy of our proposed algorithm is verified numerically and experimentally, which shows that our algorithm can be used to track the target trajectory effectively, and the error in extracting the target velocity is below 2%.