Skip to main content

Research items

  1. Detecting the nearby frames with the defect in view.

    1. We can make use of the fact that every frame captured at the same distance value for an anotated defect should have the defect in view.
    2. We can grab frames containing the same distance value going backwards from the annotated frame. (For example, say we have an defect annotation at 10.4m. We can rewind till the distance value goes lower than 10.4m and include all the frames inbetween as positive training samples)
    3. One suggested method is attempting to use OpenCV MatchTemplate funtion to essentially match the distance field with previous frames.

  2. Alternative OCR methods
    ChatGPT answers

    Some examples of video OCR software include Captricity, VideoOCR, and Pivothead.

    There are several academic institutions that have developed research-based tools for text extraction from videos. Here are some examples:
    Video OCR: Video OCR is a tool developed by the Computer Vision Lab at the University of Maryland. It is a video analysis software tool that can extract text from videos by analyzing the frames in real-time. The tool uses computer vision algorithms to identify and extract text from video frames.
    VideoTextSpotter: VideoTextSpotter is a text detection and recognition tool developed by the University of Oxford. It can detect and recognize text in videos in real-time and is capable of handling different font styles, sizes, and colors.
    ViTAS: ViTAS is a tool developed by the Image Processing Lab at the University of Valencia. It is an automatic text recognition system for videos that can detect and recognize text in multiple languages.
    TRECVID: TRECVID is an annual benchmarking activity for multimedia information retrieval that is organized by the National Institute of Standards and Technology (NIST). The challenge involves several tasks related to video analysis, including text detection and recognition in videos.
    These tools may offer more advanced features and better accuracy than some of the commercially available software tools. However, it is important to note that academic tools may require some technical expertise to use, and may not be as user-friendly as commercially available software.