We describe the development of automated emission line detection software for the Fiber Multi-Object Spectrograph (FMOS), which is a near-infrared spectrograph fed by 400 fibers from the 0.2 deg(2) prime focus field of view of the Subaru Telescope. The software, FIELD (FMOS software for Image-based Emission Line Detection), is developed and tested mainly for the FastSound survey, which is targeting Ha emitting galaxies at z similar to 1.3 to measure the redshift space distortion as a test of general relativity beyond z similar to 1. The basic algorithm is to calculate the line signal-to-noise ratio (S/N) along the wavelength direction, given by a 2D convolution of the spectral image and a detection kernel representing a typical emission line profile. A unique feature of FMOS is its use of OH airglow suppression masks, requiring the use of flat-field images to suppress noise around the mask regions. Bad pixels on the detectors and pixels affected by cosmic rays are efficiently removed using the information obtained from the FMOS analysis pipeline. We limit the range of acceptable line-shape parameters for the detected candidates to further improve the reliability of line detection. The final performance of line detection is tested using a subset of the FastSound data; the false detection rate of spurious objects is examined by using inverted frames obtained by exchanging object and sky frames. The false detection rate is < 1% at S/N > 5, allowing an efficient and objective emission line search for FMOS data at the line flux level of greater than or similar to 1.0 x 10(-16) erg cm(-2) s(-1).