Deep neural network-based human object detection systems have become core technologies in fields such as autonomous driving, intelligent surveillance, and security detection. However, with the gradual maturation and widespread application of these models, security issues have become increasingly prominent, particularly the threat of physical adversarial attacks on object detectors. This paper proposes a novel method to enhance the transferability of adversarial attacks through ensemble feature distortion and gradient variance reduction. By suppressing significant features and enhancing background features in the feature maps of ensemble detectors, the method induces object detectors to disrupt common features. Meanwhile, a gradient variance reduction-based ensemble strategy is introduced to prevent overfitting and improve the transferability of adversarial attacks. Experimental results demonstrate that the proposed method shows effective adversarial performance across multiple detector architectures and enhances transferability between different detection model architectures, outperforming other physical adversarial patch attack methods.