#image recognition
Traffic Camera Mistakes Woman for Car, Issues Ticket to Car Owner
Maybe we should wait some time before we apply image recognition software in more critical situations…?
We switched from the new software to working on updating K-9’s hardware over the weekend.
- Replaced the heavy SLA batteries with LiFe batteries. This reduced K-9’s weight from about 44 lbs down to 38 lbs with the same battery capacity.
- Re-wired the 12v circuit using a 24v to 12v converter as opposed to pulling the 12v off of just one of the batteries. This avoid discharging one battery faster and damaging the battery.
- Started the circuitry for the 24v battery monitor that will provide battery status on the remote. Just need to mount this on the main electronics board.
- Started the circuitry for dimming the hover lights and eyes. We will have to solder in a few more wires to complete this and mount it to the main electronics board.
- Cut a hole in front of the speaker and covered with the a grate to enhance the volume.
Once we get all of the new circuitry wired, we can test the updated software.
We also did some research on AI visual object detection and may order a Google Coral co-processor to play with. The Coral AI processor is designed to run Tensor-flow models for AI tasks like object detection and facial recognition. With K-9’s camera and the Coral board, we could have a game of playing with K-9 where you hold up and object and he recognizes it and speaks its name. The demo object database has about 1,000 objects. This would make K-9 look pretty smart and would be incredible to implement!
Finally saw Mitchells vs the Machines so:
ThisCLIP-based method does better I think in part because it can only pick from among the tags I give it.
Visual chatbot, never change.
