結合雙AI晶片與熱成像溫測模組之自動目標搜索與溫度量測系統

Abstract

本論文提出一種自動搜索目標系統，使用雙人工智慧邊緣型運算處理器結合紅外線熱成像感測器，並透過步控制進馬達來實現自動搜索目標且掃瞄範圍擴增的人體溫度測量設備。本文首先回顧深度學習及類神經網路對於影像辨識的起源以及其應用性，並探討邊緣型處理器對於人形偵測的可行性，再根據此基礎發想出測量人體溫度之應用。而後介紹本論文主要系統架構及硬體設備，使用Mipy深度學習AI開發板配合多種感測裝置，來達成AI目標辨識及環境訊息的測量。本系統架構建立於模型本身的可靠性，針對模型訓練的部分有加強描述：從目標圖片的選取及拍攝、訓練過程的流程改善及參數調整、及最後模型在實驗環境的誤判修正。接著將訓練好的模型載入雙Mipy深度學習AI開發板，並制定一套演算法，協調各微處理器間的交互關係，達成快速掃描且穩定測溫的功能。最後針對多個實際場景，驗證本論文所描述之目標以及該架構反應速度與正確性。

This thesis presents an automatic target searching system using dual AI chips combining an thermal imaging module and stepping control to implement a temperature measuring device that can automatically search for targets and increase the scanning range. This thesis firstly reviews the origin and application of deep learning and neural network for image recognition. We explored the feasibility of the edge processor for human detection. We developed the application for measuring human temperature. This thesis introduces the main system architecture and hardware devices. Mipy development board combines various sensors to complete AI target recognition and collection of environmental information. The system framework is built on the reliability of the model, so we enhance the description for model training, such as the selection and capture of the target, the improvement of the training process and the adjustment of parameters, and the correction of model false positives in the experiment. We load the trained model into the dual AI development board and develop an algorithm to coordinate the interaction between the microprocessors to achieve fast scanning and stable temperature measurement. Finally, the application was validated against several practical scenarios to accomplish the objectives described in this thesis, and the framework was responsive and effective