Information regarding the precise start and end of a television show is critical for supporting user journeys based on shows beginning or ending, such as tuning in precisely when a show begins, time-shifting the broadcast for later viewing, etc. However, the accuracy of broadcast TV schedules is limited. Scheduling metadata are neither consistently available nor accurate. This disclosure describes techniques to detect precise start and end times of a TV show by comparing fingerprints of video frames when it airs across different channels. Video chunks retrieved from a content directory are converted into a series of video frames and fingerprinted. Fingerprints are compared across channels to find a matching frame in one channel within a threshold distance of a frame in a second channel. Secondary comparison is performed to identify and correct misalignment resulting from differences in content transitions across channels. Optionally, video frames are cropped before fingerprinting to remove portions containing variable content such as news tickers. The identified first and last set of video frames that match across two or more channels serve to identify the precise start and end of a TV show, respectively and are used to support functionality that is dependent on start and end times.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Kimball, Jason, "Video Frame Fingerprint Matching for Precise Detection of Show Start and End Times", Technical Disclosure Commons, (September 07, 2023)