Abstract: Taking the infrared early warning system for detecting and identifying the terminal-sensitive projectile (TSP) as application background, thermal camera was used to capture and track the TSP which was thrown from a hot air balloon in real time, and the infrared image sequences in complex cloud layer background of TSP in steady-state falling were obtained. Since infrared images had the features of fuzzy, low SNR et al, a detection algorithm of combination spatial filter was introduced. In this method, the wavelet threshold and median filter algorithm were combined to denoise, the Top-Hat transform was presented to restrain the background, and target feature was extracted and the grayscale contrast between the target and the background was enhanced with the morphological filter. The multi-frame detection results show that the combined filter algorithm can effectively detect the TSP from the experimental image sequences, thus providing basis for further technology research of target tracking.