Against the backdrop of rapid urbanization and motorization, left-turn motor-non-motor vehicle conflicts at urban four-phase signal-controlled intersections have become increasingly prominent, seriously undermining traffic efficiency and safety. This study uses UAV aerial photography to collect traffic data during morning and evening peak hours, extracts key vehicle trajectory parameters, and focuses on three core directions: conflict characteristics, quantitative models, and mixed traffic optimization. It clarifies conflict definitions and classifications, systematically analyzes the expansion effect of left-turn non-motor vehicles and four core conflict scenarios, derives the calculation method for the maximum expansion width of left-turn non-motor flows, verifies their normal distribution characteristics, and builds well-fitted expansion width and conflict frequency models via multiple linear regression. Finally, it classifies mixed traffic into four categories using K-means clustering and proposes targeted intersection optimization measures, providing a quantitative basis to alleviate conflicts and enhance overall traffic quality.