結合影像與四輪控制之智能車體與人體追蹤系統

Abstract

本研究設計一套可以辨識並追蹤人體的智慧型四輪驅動車。結合了四輪驅動車與影像辨識模型，車體由四顆馬達驅動四個車輪，具備靈活的運動控制能力，達成穩定追隨目標的控制。在影像辨識上，車體搭配深度攝影機擷取前方人體影像，透過關節偵測工具MediaPipe Pose擷取人體背部關節的移動資訊，並使用GRU時間序列模型學習並記憶主人的關節動作模式。當系統辨識出該模式與記憶資料相符時，會將此人標記為主人（Master），並啟動自動追隨模式。在運動控制方面，利用運動方程式得出的轉速控制馬達，並透過增量型PID控制器調整轉速，讓車子轉彎與前進更穩定。此研究整合了影像辨識、深度學習與四輪控制技術，有效提升四輪驅動車對指定目標的辨識準確率與追蹤穩定性，未來可應用於智慧陪伴、物流配送及購物跟隨等場域。This research designs an intelligent four-wheel mobile platform capable of recognizing and tracking humans. By integrating a four-wheel drive vehicle with an image recognition model, the system utilizes four motors to drive the wheels, providing flexible motion control to achieve stable target tracking.For image recognition, the vehicle uses a depth camera to capture the person in front. Using MediaPipe Pose, it extracts movement information from the person’s back joints. A GRU time-series model learns and memorizes the Master’s joint movement patterns. When the system detects a matching pattern, it identifies the person as the Master and activates autonomous following mode.In terms of motion control, the vehicle calculates motor speed using motion equations and applies an incremental PID controller to adjust the speed, ensuring smoother turning and forward motion.This research integrates image recognition, deep learning, and four-wheel control technologies to enhance the recognition accuracy and tracking stability of the four-wheel drive platform. It holds potential for future applications in smart companionship, logistics delivery, and shopping assistance scenarios.