Background radiation noise interference is a difficult technical problem for dim signal detection. A dim target detection algorithm was proposed which can significantly improve signal-to-noise ratio(SNR) to achieve uniformly motion dim target detection successfully. Firstly, a coordinate space and a velocity space were established. Then the original image sequence was stacked along different velocity vectors to acquire a new image sequence with SNR improved and the new image sequence forms an image space. Secondly, quasi-target points in the image space were detected by constant false-alarm ratio(CFAR) judging. Finally, velocity vectors and coordinate vectors of quasi-target points were mapped to the velocity space and the coordinate space respectively. As a result, two local peaks from the spaces will confirm true target points. Experiments of real images from actual IR imaging system show that proposed algorithm can improve SNR approximately up to n^1/2 times of original image SNR, and the proposed algorithm is demonstrably superior to compared algorithms on detection probability and false alarm probability.