Rendering a scene with participating media is done by generating a depth map from a camera viewpoint and a shadow map from a light source, converting the shadow map using epipolar rectification to form a rectified shadow map (or generating the rectified shadow map directly), generating an approximation to visibility terms in a scattering integral, then computing a 1D min-max mipmap or other acceleration data structure for rectified shadow map rows and traversing that mipmap/data structure to find lit segments to accumulate values for the scattering integral for specific camera rays, and generating rendered pixel values that take into account accumulated values for the scattering integral for the camera rays. The scattering near an epipole of the rectified shadow map might be done using brute force ray marching when the epipole is on or near the screen. The process can be implemented using a GPU for parallel operations.