Traditional lower limb prosthesis motion intent recognition often uses multi-modal sensors to collect time-frequency domain features of the residual limb side. In short-term intent recognition, it has certain complexity and lag, and time-frequency domain features are unstable, which is difficult to achieve the purpose of real-time intent recognition. This paper proposes a real-time intelligent lower limb prosthesis motion intent recognition method based on improved template matching technology. After the amputee's motion pattern is acquired, onlys an inertial sensor is used to collect the data of the sound limb side in the swing phase. Based on the improved template matching, a complete template library is created through sliding window, so that each kind of motion pattern is sufficient atomic patterns in the library, and the motion intent of the lower limb prosthesis is recognized in real time. The experimental results show that the recognition rate of the algorithm reaches 99.50% in 5 steady states: level-ground walking, stair ascent, stair descent, ramp ascent and ramp descent, and 97.03% after introducing 8 transitional states. This method can greatly improve the real-time recognition performance of lower limb prosthesis and help unilateral lower limb amputees walk more naturally.