A multi-model approach for distance and angle estimation using a custom-designed tag

Recent advancements in deep learning-based object detection models have led to remarkable developments across various application domains. However, the accurate estimation of distance and angle information, alongside object detection, holds critical significance for applications like autonomous vehicles, industrial processes, and remote sensing technologies. This study proposes an algorithm employing multiple models to address the problem of detecting the distances and angles of objects. The algorithm is developed to precisely determine both distance and angle information of a custom-designed tag. The algorithm employs a two-stage structure, with the first stage utilizing two distinct CNN models to detect the tag and its components. In the second stage, the data obtained is analyzed through a MLP to predict distances and angular values. The performance of the proposed algorithm has been evaluated through experiments simulating real-world applications, yielding reliable results. This work seeks to advance applications that integrate object detection with precise angular measurements. In experiments conducted under four different test environments simulating real-world conditions, the system demonstrated high detection accuracy with an average valid measurement rate of 86.72%. Additionally, the MLP model used for distance and angle estimation achieved an $R^2$ score of 0.983 on the test data. With a total processing time of approximately 0.2 s, the algorithm also shows strong potential for real-time applications. This study aims to contribute to applications that require integrated and reliable estimation of object position and orientation alongside object detection.