Indoor robots are often constrained by environmental complexity, densely distributed obstacles, and path oscillations in autonomous navigation, and they are in urgent need of efficient and accurate navigation algorithm support. Based on this demand, a visual navigation system integrating the dynamic window method and the improved firefly algorithm (DWA-FA) is constructed, combining the global guidance mechanism with adaptive multi-objective optimization to improve the global consistency of path planning and the local obstacle avoidance ability. Experimental results show that the planning success rate of DWA-FA reaches 98.5%, which is 13.5% higher than that of FA-DQN, the average path length is shortened to 5.64 m, and the planning calculation time is reduced to 0.87 s. In terms of obstacle avoidance performance, the obstacle miss detection rate is reduced to 1.3%, the dynamic response delay is only 74 ms, and the path deviation is controlled at 0.09 m, showing high stability and accuracy. DWA-FA can provide an effective solution for efficient and accurate robot visual navigation in complex indoor environments.