Moving Object Detection Using The Yolo Method And Video-Based Depth Estimation

Abstract

The development of artificial intelligence (Artificial Intelligence), particularly in the field of computer vision, has driven advances in image and video processing to automatically understand environmental conditions. One of its applications is object detection; however, detection methods generally only provide information about the type and location of objects within bounding boxes without information about their distance from the camera. This limitation means that the information generated is insufficient to support systems requiring a more comprehensive understanding of the environment. Therefore, an approach that combines object detection and distance estimation using an RGB camera without additional hardware is required. This study aims to develop a video-based object detection and distance estimation system using a combination of YOLOv10 and ZoeDepth. YOLOv10 is used to detect objects and determine their positions in each video frame, while ZoeDepth is used to perform depth estimation to obtain the estimated distance of objects from the camera. The research methodology includes data collection, preprocessing, model implementation, and system evaluation. The data consist of 250 CCTV video samples, comprising 162 secondary data and 88 primary data. All data were used as testing data because YOLOv10 is a pre-trained model previously trained using the COCO dataset. Object detection performance was evaluated using the mAP metric, while distance estimation was evaluated using Root Mean Square Error (RMSE), with the iBims-1 dataset serving as the ground-truth depth reference. The test results show that the system is capable of detecting several objects, namely cars, buses, trucks, and persons, while also providing distance estimates for the detected objects. The object detection evaluation achieved an mAP value of 81.03%, while the distance estimation evaluation using ZoeDepth produced an RMSE value of 4.2248 meters. Based on these results, the combination of YOLOv10 and ZoeDepth successfully produces a system capable of performing object detection while simultaneously providing video-based distance estimation using an RGB camera. This system can serve as an approach for developing computer vision applications that require both object and distance information. This approach can improve the completeness of visual information to support automated environmental monitoring.