Reliable Video Clock Time Recognition

We propose a novel approach to read the video clock in real time by recognizing the clock digits using a few techniques relative to the transition patterns of the clock. With these techniques, the clock digits are located without recognizing all the text characters overlaid together with the clock on the video, the sample digit patterns are automatically extracted and labeled in real time. The clock digits are recognized while the video is playing back without any offline training. The recognition result is further verified against the digit transition sequence to improve the performance. Experimental results show that this approach is reliable and the result is accurate regardless of the variation of font, size and color of the digit characters in different videos.