A light field camera is used to collect particle images, and the tomographic inversion algorithm is improved to obtain the three-dimensional spatial position of particles faster and more accurately. Based on imaging principle of the light field camera, a forward tracing model of light emitted by particles is established. On the basis of this model, a tomographic inversion model is constructed to reversely trace rays emitted from non-zero pixels and realize the mapping between non-zero pixels and spatial voxels. A descending dimension method is proposed to calculate the weighting matrix. The depth accuracy of MART algorithm is improved by combined with the similar triangle principle. According to the Gaussian Blob model, center position of voxel with the strongest intensity is taken as the three-dimensional position of particles. Experiments indicate that the improved MART algorithm can significantly reduce calculation time and memory size while meeting the accuracy requirement. The x-axis coordinate error is ±0.16 mm. The y-axis coordinate error is ±0.18 mm. The z-axis coordinate error is ±1.8 mm. It is more suitable for occasions that demand higher calculation speed and has better practical application value.