Abstract: A flow velocity is one of the most important physical parameters to quantify the hydraulic characteristics of water flow. The accurate measurement of shallow flow velocities can greatly contribute to understanding and simulating the sediment transport and soil erosion. In this study, a novel observation system of thermal infrared imaging and computer vision was established to measure the velocities of shallow flow. The observation system consisted of three subsystems, such as the thermal tracer control, image acquisition, and transmission, as well as image calculation. Specifically, the control subsystem of the thermal tracer was mainly responsible for the constant temperature of hot water, including the electric heater, temperature sensor, and hot water pump. The subsystem of image acquisition and transmission consisted of a FLIR ONE 3.1.0 thermal infrared camera, 4 thermal infrared targets, and a wireless router, particularly for the thermal infrared images of thermal tracer migration. The subsystem of image calculation was used to extract the high frame rate from the thermal infrared images, including the data storage and computing matching. As such, the velocity of shallow flow was dynamically monitored using the automatic control of thermal tracer, instantaneous image acquisition, image correction, noise removal, and centroid determination. Furthermore, a series of experiments were conducted to verify the system in the Simulated Runoff Hall of the State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau. The experimental glass tank was in the size of 4.6 × 0.196 × 0.1 m3 at a gradient of 15° to the horizontal, together with the different flows (0.2-1.5 L/s). The results showed that the observation system presented a higher accuracy than before, suitable for the dynamic transport of thermal tracer in the different temporal and spatial scales. Specifically, an excellent performance was achieved, where the measurement standard deviation of the system was 0.0201 m/s, the observation accuracy reached 98.33%, the observation time resolution was 1/9 s, and the spatial resolution was up to 2 mm. Moreover, the accuracy of the observation system with the thermal infrared imaging was much higher than that with the traditional tracer techniques (dye and salt tracer). The maximum and minimum relative errors of the observation system were –9.61% and 0.16%, respectively, and the range of the relative errors was within ?10%. The maximum relative error of the dye tracer was 75.92%, and the minimum relative error was 12.03%. The maximum and minimum relative errors of salt tracer were –26.67% and 0.11%, respectively, where 52% of the samples presented the relative errors within ?10%. Correspondingly, the observation system with thermal infrared imaging was provided a reliable way to measure the shallow flow velocity. By contrast, either the dye tracer or salt tracer cannot calculate the overall situation of water flow, due to the manually recording one-time value during measurement. Fortunately, the observation system with thermal infrared imaging can be widely expected to accurately record the overall situation of the water flow. More importantly, the leading edge points and centroid points of the thermal tracer area can be identified at different time intervals, thereby accurately calculating the leading edge velocity and centroid velocity of the water flow. The images can also be used to trace the development of water flow in various tracer sections at different times. As such, it can be possible to measure the tracer migration velocity along the direction of water flow and the diffusion velocity perpendicular to the direction of water flow in the future. This technique can also be applied to monitor the rainfall erosion and runoff scour in the process of soil erosion.