即時辨識移動物體之連續波雷達信號分析系統;Continuous Wave Radar Based Signal Analysis System For Real-time Recognition Of Moving Objects

[[abstract]]移動目標的辨識對於社會安全或是智慧生活產品上有很大的幫助，能在災區協助救災人員尋找生命，也能透過步態分析在醫療上能夠提早發現疾病，然而目前大部分研究是利用攝影機蒐集影像進行影像特徵提取與分析來達到追蹤與辨識。本研究使用連續波雷達，反射回來的雷達波能根據都卜勒效應計算出目標運動的方向與速度。由於不同物體移動時會有各自的行為特徵，透過時頻分析其週期變化可提取有用的特徵進行分類與辨識，我們的目的是要能辨識出人與非人(狗、車與空白背景)。經過了一系列的實驗，蒐集了人與非人的步態資料，我們提出四種特徵組合的方法，探討其辨識率，並透過K-nearest neighbor與Support vector machine兩種分類器，比較是否要先用Fisher linear discriminant analysis來分割。最後建立了一個簡單的使用者介面，方便進行即時的移動物體分類，來輔助未來以連續波雷達辨識物體的應用。 The recognition of moving targets is very helpful for social security or smart living products. Most research achieve the goal of objects tracking and identification by using a camera to collect image data for feature extraction and analysis. In this work, we use a continuous wave radar, and calculate the direction and speed of an moving object from its reflected radar wave according to the Doppler effect. Since different objects have their own motion behavioral patterns, we can perform classification or recognition tasks by analyzing its periodicity with time frequency analysis and extracting meaningful features. Our goal is to differentiate human and non-human objects like dogs, cars, and blank background.After collecting the human and non-human gait data through a series of experiments, we proposed four combinations of features, and used K-nearest neighbor and Support vector machine classifiers to check if Fisher linear discriminant analysis is required as a preprocessing. The recognition rate under different feature combinations and classifiers were compared. We also establish a simple user interface to facilitate the real-time moving target recognition for future applications