Using neural network for motion analysis

K. C. Chan, J. Xie, B. Shirinzadeh

Research output: Chapter in Book/Report/Conference proceedingConference PaperOther


An algorithm for real-time object tracking using Artificial Neural Network (ANN) is proposed in this paper. This algorithm consists of an ANN for determining the exact position of the moving target and a next position prediction algorithm. The basic theory of this algorithm is using two processing windows to follow the object being tracked. By analyzing the relationship between the area of the object enclosed by the windows and the relative position of the object within the window, the position of the object in the world frame can be obtained. A three layer ANN was built and trained to learn this mathematical function. The next position prediction is achieved using an approximated jerk. This algorithm predicts the movement of the moving object and then move the windows to the predicted position. The ANN is then recalled to calculate the accurate object position in real time. The theory of the algorithm in conjunction with experimental validation will be presented in this paper. The successful use of Neural Network in this new algorithm for 2-D motion analysis provides an efficient and practical approach for solving the 2-D object tracking problems in the manufacturing area.

Original languageEnglish
Title of host publicationProceedings of Control 92 -Enhancing Australia's Productivity Through Automation, Control and Instrumentation
Number of pages5
Edition92 pt 15
Publication statusPublished - 1 Dec 1992
Externally publishedYes
EventControl 92 -Enhancing Australia's Productivity Through Automation, Control and Instrumentation - Perth, Australia
Duration: 2 Nov 19924 Nov 1992

Publication series

NameNational Conference Publication - Institution of Engineers, Australia
ISSN (Print)0313-6922


ConferenceControl 92 -Enhancing Australia's Productivity Through Automation, Control and Instrumentation

Cite this