Real-Time Object Analysis : - Using a Three Dimensional Camera

Real-time object detection, tracking and movement prediction (Real-time object analysis)using Three-Dimensional (3D) cameras represent a critical advancement in computervision with applications ranging from autonomous systems to industrial automation. Thisproject investigates the integration of stereo, RGB vision and machine learning to create amodular component capable of detecting objects, estimating distances and a simpletrajectory prediction in real time. The motivation stems from the need for adaptive systems,capable of perceiving dynamic environments accurately and efficiently. To address this, wedeveloped an artifact using the OAK-D Lite stereo camera, leveraging You Only LookOnce (YOLO) version 11 small for object detection, ByteTrack for multi-object trackingand DepthAI’s spatial location calculator for 3D localization. Our solution processesRGB-D tracking data to compute simple object movement predictions and outputs allresults in a structured Javascript Object Notation (JSON) format for integration with othersystems. Experimental results demonstrate solid performance with distance estimationerrors under 15% for objects within 10 meters and real-time processing at 15–30 FramesPer Second (FPS). The artifact’s design follows the Design Science Research Methodology(DSRM), emphasizing practicality for edge deployment. This work contributes a reusablecomputer vision component for adaptive systems while highlighting challenges in depthaccuracy and occlusion handling.