Object Tracking using APRON (with CUDA extensions) This video shows the APRON software performing tracking of a "learned" object. Initially, the object of interest is captured and learned. Then the object is tracked on a live-video feed from the webcam. The performance is about 30fps (although the video shows 20fps due to screen-recording), which is about the limit of the webcam, indicating it could go much faster. The learned object and the video image are converted into the frequency domain, and the convolution of the two produces a similarity map. An interesting side-effect of this is the algorithms robustness to orientation, as highlighted in the video. CUDA is used to perform the transform and do the convolution, through CUDA extensions built into the APRON simulator. The algorithm however is programmed entirely in APRON Script. APRON Demonstration Video This video was originally just a test to see if I can record the screen and upload a meaningful video to YouTube (all of which was a bit new to me!). The video shows several aspects of the APRON simulation environment, whilst a Locally Adaptive Thresholding algorithm is run.

All material contained on this website is copyright of David R W Barr 2009