From Far-Field Dynamics to Close-Up Confidence: Action Recognition Across Varying Camera Distances