With the expansion of computer vision, obstacle avoidance systems have gained significant priority. Although robust algorithms exist for this task, the cost is high due to the stereo camera setup. The proposed approach integrates Farneback’s method for optical flow analysis with a monocular depth estimation model using MiDaSv2.1 for relative depth perception. The depth map output helps estimate the proximity of objects, and the optical flow helps to detect the movement of a dynamic object across consecutive frames. A dynamic threshold adjusts according to the perceived environment, allowing the system to detect potential obstacles in various real-world scenarios. The average individual frame processing time is 0.27 s, ensuring the system is effective in real-time processing. Our approach gives 79.1% accuracy and 79.7% recall on a custom dataset. The accuracy metric is affected by the relative nature of monocular depth estimation, leading to an increase in the count of false positives. A key advantage of our system is its ability to operate on CPUs, which further cuts down on hardware costs by eliminating the explicit need for GPUs.