Running-related knee injuries are highly prevalent and are closely associated with excessive or poorly managed knee loading. Recent injury prevention strategies emphasize running posture and load management rather than focusing solely on local anatomical structures. In this study, a non-wearable, image-based system is proposed for detecting peak knee loading during running based on key running pose detection, providing a potential alternative to conventional hardware-based assessment methods. A geometric knee-loading region, referred to as the knee–hip–ankle (KHA) region, was defined by connecting the hip, knee, and ankle landmarks of the stance leg, and its area was computed on a frame-by-frame basis to characterize image-based knee-loading patterns. Peak knee loading was operationally defined within the Pose Method framework as the instant when the stance-leg knee joint angle reaches its minimum value, which was used in this study as a theory-driven image-based indicator for locating the key running pose. A biomechanical experiment involving six participants was conducted to examine the effects of running speed and gradient on peak knee-loading characteristics. The results showed that higher running speeds and steeper gradients were associated with enlarged KHA region areas, suggesting increased knee-loading characteristics under greater running demands. These findings demonstrate that the proposed system can capture meaningful changes in peak knee-loading characteristics and may support running posture evaluation and knee load management.