In 2004, I attended a talk given by Muneaki Ohshima, Ning Zhong and Shinchi Murata titled Peculiarity Oriented Analysis in Multi-People Tracking Images.
In the talk, they shared their application of a peculiarity mining approach to analyse image sequences of people walking.
They noted that peculiar data are a subset of objects in the database and are characterized by two features:
They are very different from other objects in a data set.
They consist of a relatively low number of objects
In the same year, I visited Janez Pers in his laboratory at University of Ljubljana to discuss his work on tracking players in sport games using computer vision.
Both research groups had fascinating insights to share and raised fundamental questions to inform the pattern recognition theme in our Sport Informatics and Analytics course.
Muneaki, Ning and Shinichi used video recordings of people’s movement behaviour in and around the ticket hall at a railway station.
In order to explore each person’s movement patterns, they converted the raw video data into XY coordinates and a CSV format was applied to each video frame. Each person was allocated a unique ID and each video frame was numbered.
Muneaki, Ning and Shinichi hypothesised that most people in the ticket hall would walk in a linear way towards their objective and they proposed:
whether the behaviour of the person is usual or not can be analyzed by calculating the segment number (i.e. the number of changes of direction) in the linearized walking data of each person.
Their paper shares the process of how they developed their detection algorithm and the modifications they made as they interrogated the data available to them.
I became aware of Janez Pers interest in player tracking in 2000. At that time, he had co-written a paper with Stanislav Kovacic that reported the development of a computer vision system for tracking indoor games with the game of handball as the focus of their attention.
The system used two fixed S-VHS cameras above the handball court. The raw video was transformed into digital format by the use of a Motion-JPEG frame grabber at 25 frames per second.
As with Muneaki, Ning and Shinichi’s discussion of the changes to their system, Janez and Stanislav reported changes they made to their approach to data acquisition.
Despite using an established technique to calibrate the cameras, they decided to address a fundamental issue with radial distortion. Their new approach corrected the distortion. The correction can be seen in this picture.
Janez and Stanislav reported:
We developed three different algorithms for use in player tracker: motion detection, template tracking and color based tracking. We tested different combinations of these algorithms; each one has its own advantages and disadvantages.
They concluded from their handball data:
The system is capable of outputting the positions of every player in court coordinates in every instance of time. It was used to track players in the handball match to perform analysis of their movement. It forms a base of the complete sports analysis system, which will include advanced data processing, statistical analysis and presentation capabilities.
Muneaki, Ning, Shinichi, Janez and Stanislav provide an introduction to some of the fundamental ideas in computer vision. I have chosen their two papers as examples of transparent accounts about how to combine theoretical insights with empirical investigation.
They have continued their work in the field. A decade after their 2004 paper, Muneaki and Ning were working on the analysis of EEG data.
In 2015, Janez and Stanislav were part of a research team investigating visual re-identification across large, distributed camera networks.
In June 2017, Computer Vision and Image Understanding published a special edition dedicated to computer vision in sport.
The Third International Workshop on Computer Vision in Sports was held in July 2017. Papers from the first international workshop in 2013 were published in 2014 and that publication was promoted as the first book of its kind devoted to the emerging field of computer vision in sports.
