Distance Estimation of Vehicles using Triangle Similarity and Feature Extraction
Mary Christine Tomas, Joshua Dave Oropilla, Elijah Nathaniel Senga, Daniel Tandoc · 2023
The present study seeks to improve upon the study conducted, with the researchers employing a different depth estimation computation method and an enhanced object detection model. The constituents will use their captured data with the OpenCV library to train the Model, whose architecture is YOLOv5. The constituents have also contributed data where the camera view is obstructed by natural illumination or weather to determine whether the Model also detects vehicles in an obstructed view. Since the previous study relied on trial-and-error methods to estimate depth, the Model's constituents will employ the Triangle Similarity method [7]. The Model is operational because of the input data passing through the YOLOv5 Architecture, head, neck, and back. Once the vehicle is identified, its focal point will be used to ascertain its distance, continuously updated in real-time. Since YOLOv3 was used as the Model for object detection in the previous study, using the YOLOv5 architecture has improved the accuracy, according to the constituents. Using Triangle Similarity, the previous study's trial-and-error technique for depth estimation was improved. Given that the previous study relied solely on trial-and-error methods for depth estimation, this study may have room for improvement [14].