Approach for Automatic Text Detection and Localization in Video Frames

People who mess with security cameras in the middle of a video frame delete the video evidence. The courts and police need to change how they do their jobs. It can be challenging to tell the difference. A lot of people enjoy these games. A lot of different hand- made ways have been found to do it over the years. Of course, it takes a lot of work to make tools that can quickly and correctly figure out what kind of hacking it is, call it by its name, and explain it. CNNs that work in two dimensions can help deep automatic feature extraction do their job if they know what happened and when it happened. We want to try this new idea. First, we use an autoencoder to separate the feature space into smaller pieces that are easier to handle on the computer. Next, we find links between video frames that are very far apart. We can find hacker signs with this. More than 90% of the time, this method finds places where people can cheat. There is no difference in the style, frame rate, quality, or number of frames that change.