CS 365 : Artificial Intelligence

Tennis Stroke Detection

---

Final Presentation
Final Report
Code Tarball*

*Data could not be loaded along with code due to its large size.

Proposal :

Human activity recognition has been studied in many research domains, and sports video analysis is one of them. This project proposes a method of stroke detection in tennis videos basically using particle filtering techniques. Stroke Detection is based on a combination of particle filters, motion descriptors and event detectors. Firstly, we track the player and extract player centered images. Second, we use the iterative application of the famous Lucas Kanade algorithm for analyzing the optical flow. And finally, we generate motion descriptors. The task of feature detection has been done using an extension of the Voila Jones algorithm in 3-D. Adaptive Boosting algorithm of machine learning has been used for the purpose of training.

Our task can be divided into 4 subtasks :

• Player Tracking using particle filters
• Optical Flow Analysis: We have used Lucas Kanade Algorithm of optical flow analysis in this project.
• Feature Detection: We have used a 3-D extension of the famous Voila Jones algorithm.
• AdaBoost: We have used adaptive boosting algorithm of machine learning for the purpose of training.

Player Tracking with Particle Filter:

Sparse Optical Flow:

3-D Haar-like features:

Adaboost Classifier:

Papers Referenced :

• An Adaptive Color Based Particle Filter. Katja Nummiaro, Esther Koller Meier, Luc Van Gool
• P. Voila and M. Jones: Robust real-time object detection. In Proc. of IEEE Workshop on Statistical and Computational Theories of Vision [2001], pg. 4-8.
• Particle Filtering and Object Tracking by Rob Hess [2006]
• Jean- Yves Bouguet: Pyramidal implementation of the Lucas Kanade Feature Tracker, Description of the algorithm [2001], pg. 2-4.
• Klas Nordberg and Gunnar Farneback: A Framework for Estimation of Orientation and Velocity, Computer Vision Laboratory, Department of Electrical Engineering, Linkoping University[2001],pg. 3-4.
• Efficient Visual Event Detection using Volumetric Features by Yan Ke, Rahul Suthankar, Martial Herbert [ICCV’05], pg. 2-4.

---