Autonomous mobile robots need precise environment models, often obtained through the use of SLAM-based (Simultaneous Localization and Mapping based) occupancy grid maps, for safe and reliable path planning. Despite the popularity and efficiency of heuristic planners such as A* search, their performance degrades with incomplete and outdated geometric maps, especially in a cluttered and partially observable indoor environment. This paper investigates whether RGB-D (red-green-blue and depth) image data can be used to improve traditional path planning without modifying the classical A* framework. The proposed approach combines the classic A* planning algorithm with a perception module that uses a cost function C(n) determined based on visual data (based on the SegFormer-B0 vision transformer) and is compared to a baseline A* algorithm that operates using an RTAB-Map (Real-Time Appearance-Based Mapping) occupancy grid. RTAB-Map is used as a graph-based SLAM framework for generating the geometric map of the environment. In the vision-augmented systems, segmented obstacle masks are combined with the aforementioned SLAM map using a soft penalization layer that steers the search process away from visually predicted danger areas. Experiments are conducted within a ROS2 (Robot Operating System 2) with Gazebo simulated indoor environment, where the mission duration, path length, deviation from the plan, path smoothness, number of maneuvers, and collision rate are evaluated. The results show that the proposed visual technique, in the tested scenarios is capable of eliminating collisions by avoiding them, enhancing trajectory stability and improving the adherence to the desired path compared to the baseline approach. The proposed vision-derived penalization layer improves mission reliability in indoor navigation scenarios, however it introduces more conservative behavior and longer mission times. These findings indicate that lightweight transformer-based perception can be integrated with SLAM-based occupancy maps to support safer classical heuristic planning in complex indoor environments.