Optimizing image-based surface velocity measurements: sensitivity to graphical enhancement and processing parameters

Abstract

Introduction Image-based techniques are being increasingly used to monitor natural and artificial rivers due to their advantages over conventional measurement methods. These optical approaches are non-intrusive, allowing data collection without interfering with the flow. Their versatility enables application in diverse conditions, including dangerous situations such as floods, ensuring both operator and equipment safety. By exploiting high-resolution cameras and advanced image processing software, these methods can estimate surface velocity even when traditional techniques are impractical or unsafe. Despite their benefits, the accuracy of image-based measurements depends on several factors. Environmental conditions can affect image quality, leading to potential inaccuracies in velocity estimation. River characteristics (e.g., turbulence, turbidity, vegetation) further influence data capture. Additionally, the effectiveness of these methods relies on proper field acquisition protocols and precise parameterization of the processing software. Optical techniques follow a three-phase workflow: (i) seeding, (ii) recording, and (iii) processing. The seeding phase involves introducing natural or artificial tracers onto the water surface to track movement. During recording, video sequences are captured using stationary or mobile platforms (Unmanned Aerial Systems, UASs). The final phase consists of three key steps. First, the recorded video undergoes transformations to enhance quality and stability, including stabilization to correct unwanted camera movements, orthorectification for geometric accuracy, and graphical enhancement to improve tracer visibility. Next, surface velocity is estimated using correlation-based algorithms or similar techniques to track the tracers’ movement across consecutive frames. Finally, post-processing refines velocity data by filtering noise, interpolating missing information, and generating an average velocity map. Among these steps, graphical enhancement plays a particularly crucial role. By increasing contrast between tracers and background, it improves motion detection and reduces errors in velocity estimation. However, its performance does not solely depend on graphical enhancement; the parametric setup of the processing software is equally critical. Improperly tuned parameters can introduce biases, compromising results. Materials and methods This work presents a comprehensive sensitivity analysis, assessing how different graphical enhancement techniques and processing parameters jointly affect measurement accuracy. The study examined the impact of various enhancement methods and algorithmic settings on velocity estimates. The analysis focused on a range of filters, from the traditional ones (e.g., intensity capping, high pass, contrast adjustment) to advanced techniques (e.g., background subtraction, edge detection filters). The goal was to identify optimal strategies that maximize accuracy while minimizing susceptibility to errors due to environmental and technical constraints. To achieve this, a dataset of video recordings was collected during discharge measurement campaigns on two Sicilian rivers, conducted over several days to account for a wide range of acquisition conditions. Videos were captured using both UAS platforms (i.e., a small drone) and fixed cameras, ensuring a diverse range of observational conditions. Each video sequence was appropriately stabilized and orthorectified, with only the portions containing an adequate distribution and density of tracers selected for further analysis (Alongi et al., 2023). The tracers were manually introduced by operators onto the water surface to ensure consistent tracking throughout the recordings. The data were processed using PIVlab (Thielicke & Stamhuis, 2014) and SSIMS-Flow (Ljubičić et al., 2024), two widely used software tools for image-based analysis. The study also investigated the key