Aiming at the infrared object detection applications, a novel generalized cumulative sum processing is presented. Since in a typical IRST application system, object appearing and vanishing can be regarded as the change-point detection problem in Statistics. One of the effective solutions is the generalized cumulative sum processing (GCUSUM). Analyses are focused on the detection threshold value selection of GCUSUM algorithm and relations among the threshold value and false alarm rate, detection probability and signal-noise rate. The further researches extend a uniform band IRST system into the multiple band IRST system and improve the realization of GCUSUM algorithm. Results of theoretical analysis and simulation show that our modified algorithm has excellent object detection performance in an infrared image sequences from a real IRST system.