Image impulse noise removal is essential for obtaining high-quality images. A novel pixel gradients-based adaptive iterative median filter is proposed to remove image impulse noise by utilizing the principles of thermal infrared camera imaging. Firstly， the maximum pixel gradient of the original image is computed based on the camera''s modulation transfer function （MTF）， and a corresponding set of pixel gradients is established. Subsequently， the gradient weight root-mean-square error （GWRMSE） set of the original image and the corresponding pixel gradient filtered image is computed， and the optimal pixel gradient is determined as the one corresponding to the maximum value of Gaussian distribution of the GWRMSE set. Finally， the adaptive window size and number of iterations for the proposed filter are determined according to the density and complexity of the impulse noise in the image. Extensive experimental results demonstrate that the proposed filter exhibits excellent robustness in removing 8-bit and 16-bit single-channel impulse noise images. In comparison with other state-of-the-art methods， the proposed method can remove low-density random-valued impulse noise （RVIN） and salt-and-pepper noise （SAPN） in real thermal infrared camera-acquired images in real-time while preserving more than 99.5% of original pixels during the noise removal process. Additionally， for high-density SAPN removal， the proposed method achieves competitive results， demonstrating better peak signal-to-noise ratio （PSNR） and structural similarity index （SSIM） in comparison with filtering methods of faster running time and faster execution time in comparison with denoising methods of superior PSNR and SSIM. Moreover， it can recover meaningful image details even for images severely damaged by extreme SAPN （99%）.