The significance of human activity recognition cannot be overstated, particularly in the context of infrared vision applications. In this study, we delve into the realm of human activity recognition using infrared vision, employing the robust IITR-IAR dataset, which encompasses a wide spectrum of 21 distinct action classes. Our research is characterized by a rigorous exploration of three distinct approaches applied to this dataset. The remarkable outcome of our efforts is the attainment of a high-precision accuracy rate, standing at an impressive 81 percent. This achievement notably surpasses the benchmark set by the base paper, signifying a substantial advancement in the field. Our findings illuminate the potential of infrared vision in revolutionizing human activity recognition, promising heightened accuracy and performance in this critical domain. This breakthrough holds promise for diverse applications, ranging from enhanced surveillance to more effective healthcare solutions.