The incapability of processing flow velocities under low tracer density conditions is one of the limitations of the traditional Large-Scale Particle Image Velocimetry (LSPIV). This study developed a new LSPIV algorithm, Time Frequency Analysis (TiFA), to overcome such a limitation, enhance computational efficiency, and improve the accuracy of derived velocities. TiFA investigates the temporal joint distribution pattern of two velocity components at each location. By assuming that the valid velocities follow a quasi-normal distribution in the velocity time series, TiFA can quickly and accurately separate the valid velocities from background noise and outliers. The performance of TiFA was evaluated by comparing with other algorithms including Traditional LSPIV, Ensemble Correlation (EC), Large-Scale Particle Tracking Velocimetry (LSPTV), and Seeding Density Index (SDI) in an experimental hydraulic model and two field cases. TiFA showed the highest overall accuracy and lowest computation cost in data analysis, especially under low tracer density conditions. In addition, TiFA showed its unique ability of automatically filtering out velocity data from low-quality zones such as no-tracer zones and surface glare zones. TiFA also showed its potential in processing turbulent flow. In summary, the newly-developed algorithm, TiFA, has demonstrated strong capability and competence in various flow and tracer scenarios, making it a valuable candidate for future applications.