Abstract: As the penetration rate of advanced driver assistance systems (ADAS) continues to rise, incidents are expected to increase at signalized intersections due to violations, such as non-motorized vehicles running red lights and crossing streets, potentially undermining traffic efficiency and safety. To mitigate these adverse effects, a takeover behavior was investigated in this paper for the drivers of high-level assisted driving vehicles under the scenarios with red light running of non-motorized vehicles. Firstly, a takeover experiment was arranged based on a driving simulator, considering the various intersection types and conflict conditions. And then, 30 participants were recruited to develop a generalized linear mixed model ( GLMM ) to analyze the influence of intersection type and conflict type on driver takeover performance, physiological responses, and visual behaviors when faced with non-motorized vehicles running red lights. The experiment results show that drivers at an 8-lane by 4-lane intersection experience a heightened collision risk and diminished operational stability during the takeover process compared to those at an 8-lane by 8-lane intersection. Additionally, they exhibit increased cognitive load and psychological stress levels. Furthermore, under straight-to-right conflict conditions the risk of collision is significantly greater than that under straight-to-left or straightforward conflicts, presenting alongside elevated levels of psychological tension, cognitive load, and frequency of information search during vehicle takeovers. Notably, the size of signalized intersections has a relatively minor impact on driver takeover performance in straight-to-right conflict situations. These results underscore the necessity for high-level assisted driving systems to prioritize small intersection operations and straight-to-right conflict scenarios. This research provides foundational insights for developing relevant regulations and policies as well as enhancing safety designs for advanced driver assistance technologies.