Active collision avoidance methods are crucial components of vehicle active safety systems, which can effectively prevent collisions or mitigate collision-induced losses. To address the limitations of existing methods, particularly their insufficient foresight in dynamic traffic environments, this paper proposes an active collision avoidance control method based on driving intention recognition and an improved Driving Safety Field (DSF) model to enable more proactive and stable collision avoidance. First, a Hidden Markov Model (HMM) is trained using vehicle trajectory data from a public dataset to accurately identify the driving intentions of the obstacle vehicles, including Lane Change Left (LCL), Lane Keeping (LK), and Lane Change Right (LCR). Then, an improved potential field model is established, which incorporates vehicle acceleration to more comprehensively quantify the driving risk faced by the host vehicle within the DSF model framework. Subsequently, an active collision avoidance controller combining a longitudinal dual-PID braking controller and a lateral MPC steering controller is designed. This controller initiates corresponding avoidance actions based on the results of driving intention recognition and driving risk evaluation. Finally, co-simulation and hardware-in-the-loop (HIL) tests are conducted. The results demonstrate that, compared to conventional methods lacking driving intention recognition, the proposed method can initiate avoidance maneuvers approximately 1 s earlier, thereby more effectively avoiding potential collisions. Furthermore, key vehicle stability indicators, such as lateral acceleration and yaw rate, are significantly reduced during the avoidance process, indicating enhanced stability and satisfactory real-time performance. This method provides a solution for enhancing the active safety of vehicles in complex real traffic scenarios.