This paper presents a novel solution, using a vision-sensor, to a challenging control problem in cryogenic food freezing industry. This industrial application is characterized by significant variation in input food products because cryogenics-freezing technology with its inherent process flexibility is typically used in low volume/high mix applications. Current industrial controllers use PLCs for regulating the belt speed of the tunnel, which leads to conservative set-points and consequently significant operational cost and frequent over-freezing. Servo control of the process is difficult because of the complicated nonlinear dynamics of cryogenic freezing caused by phase-change, and thermal dynamics between the frozen products and the tunnel. The solution presented in this paper uses a vision-sensor to estimate the shape, size, and heat load of food products that will enter the freezing tunnel. An analysis of the sensor location and its impact on disturbance feed-forward control is also presented. Efficacies of these developments are verified in an industrial case study using a commodity webcam for capturing and processing two-dimensional streaming images, and integrating the processed information with an industrial control system using model-predictive control architecture. The proposed solution is especially attractive for the food industry because of the low-cost and non-contact features of webcam, operational cost savings through reduced consumption of cryogen, and improved quality through reduction in variation of temperature of the frozen products.