Seven months ago, I used a pretrained neural network to detect the appearance of animals on online cameras in different nature parks, and send notifications to the @WebCamWatcher telegram channel. Now I’ll tell you a little about what happened with this venture:
- The channel gained a small but constant audience, which has organized an additional @WCWfriends chat room to discuss photos caught by the “bot.”
- I added several cameras from African national parks, which enlivened the set of pictures.
- Dima Kryukov asked me to make a similar thing for cameras on Russian rivers in order to detect boats passing by. So the channel @wcw_boats appeared, although there are still almost no viewers.
- People also came in with the idea of using a similar design to search for people ala Lisa-alert, but the idea did not go beyond the conversation.
- I tried several different pre-trained neural networks and settled with YOLOv3, which works well, although from time to time it confuses cows with birds and turtles with bears, but this is not so important.
- In order to connect new camera options and manage all sorts of settings, I rewrote the code a couple of times, until, finally, I implemented a flexible system based on a config and a set of plugins.
- Hosting this whole thing on a home machine with a not very stable load turned out badly, so the bot suffered from periodic downtime. Then I met Gaiar Baimuratov, who is also interested in creating photo traps for birds, and Gaiar suggested he could host the whole thing. As a result, Gaiar added a docker kit to my system, and yesterday the bot moved to the new home.
- I published the code of the whole system and a general description of the logic on the github.
- Gayar becomes the maintainer and administrator of the entire system, so write him about your ideas and problems at @WCWfriends chat room.
- During the last seven months, the bot gathered a lot of beautiful shots, I selected several dozens that I personally liked and decided to share them with you.