Towards automated length-estimation of free-swimming fish using machine vision

Abstract

Machine vision has been widely and efficiently used for various industrial applications. However, in marine applications, this technology is still facing big challenges due to the free-swimming nature and unpredictable behavior of the fish. This paper presents new models for length estimation of small-size free-swimming fish objects using low-cost machine vision system. These models include two camera setups; orthogonal and stereo. In the orthogonal setup, one camera is used for measurement of the object's distance. This camera is synchronized with another orthogonally oriented one for the object's length measurement. The measured distance and length are then used to estimate the actual length of the object. In the stereo setup, however, both cameras are simultaneously used to measure the object's distance and length. Automatic calibration based on fixed object with known diameter is also considered in this study to measure the real value of object's length. In particular, for short camera distance, the effect of light refraction is assumed to be negligible therefore; a linear relationship is adopted to estimate actual object's length for stereo setup. This approximation is assessed experimentally and the measurement error in the object's length is found to be within the range of ± 1%.