인공신경망을 이용한 움직이는 물체의 변위 추정
Displacement Estimation for Moving Object using Artificial Neural Network
- 주제(키워드) Displacement , ANN , Moving Object
- 발행기관 아주대학교
- 지도교수 김재현
- 발행년도 2019
- 학위수여년월 2019. 2
- 학위명 석사
- 학과 및 전공 일반대학원 전자공학과
- 실제URI http://www.dcollection.net/handler/ajou/000000028839
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
Recently, a ball flight analyzers using FM-CW (Frequency Modulation Continuous Wave) Doppler radar are used to measure data such as speed, launch angle and carry distance in screen golf or baseball games. Trackman is typical ball flight analyzer. It is a product that was recognized for its ability to measure data in professional baseball games or golf tours. However, Trackman is expensive, and it has a reluctance to purchase for ordinary users. Also, the algorithms of the product are business proprietary. Study on displacement estimation of moving objects will help develop low-cost and high-efficiency ball flight analyzers, it can be used to track projectiles. In this thesis, we proposed a physical model and a statistical model to estimate the displacement of a moving object. The first proposed model is the physical model. The forces at work in a ball during flight include lift, drag, and gravity. The moving object with a high spinning speed has a different drag and lift according to its size, depth, speed, and spin rate. The physical model was designed reflecting these differences. The second proposed model is an MLR (Multiple Linear Regression) and an ANN (Artificial Neural Network) models using statistical analysis method. In real environments, FM-CW Doppler radars have limitations in detecting objects that travel more than tens of meters. Based on interpretable predictive models, the statistical analysis method can explore hidden patterns and relationships between variables. The ANN models were changed to design the ANN structures with the best performance. The data was collected using two types of golf clubs, 7 iron and Driver. The RMSE (Root Mean Square Error) was calculated to put forward the best model. As a result, the proposed ANN model has the best performance both 7 iron and the Driver club with the RMSE of 0.55 and 0.72 respectively. This research can be used not only for ball flight analysis but also for moving object analysis such as baseball in the future.
more목차
I. Introduction 1
II. Related Works 3
A. Derivation of moving objects parameters 3
1. Speed of moving object 5
2. Launch angle 6
III. Displacement Estimation 13
A. Physical analysis 15
1. Drag force 15
2. Lift force 16
B. Statistical anlaysis 21
1. Multiple Linar Regression 22
2. Artificial Neural Network 24
IV. Performance evaluation 27
A. Data measurement environment 27
B. Simulation result 31
1. Physical model 31
2. Statistical model 34
V. Conclusion 40
References 41
