This paper investigated the working scheme of a spaceborn signal processor applied to the USA Midcourse Space Experiment (MSX) for target detection and tracking. The moving target indicator in the MSX performed the trace extraction in two stages, namely, “nominator-confirmer” procedure. the first stage used a binary matched filter to “nominate” streak paths which looked like they might actually contain signals and the second one did a energy accumulation decision to “confirm” the nomination of the first stage. Based on the study of above procedure, an improved method was proposed to detect moving targets. In “nominator” stage, the velocity restraint condition was introduced to binary matched filter for reducing the probability of false alarm. In “confirmer” stage, the criteria of “minimum sample mean difference” were added to that of “maximum sample mean” for improving the probability of detection. The improved method could be widely used in complex background images. Finally, an experimental method was designed to analyze the algorithm performance of two different methods. The experimental results show that the number of streak paths filtered by improved method is lower than 50% compared with that former method, and number of clock cycles is reduced from 8.0×106 to 7.1×106. Improved method can effectively detect multi-moving targets in real time by expanding the reasonable range of threshold to nearly 20.