Work Done

I started tagging the positive and negative images needed to train the vehicle classifier from traffic camera footage — I’m still waiting on some footage from the Fifth and Craig intersection from Zina to add more images to the training dataset and then finally train the model. I was able to ssh into the Raspberry Pi (after connecting it to my Mobile WiFi hotspot) and I was able to connect the IP Camera to my Mobile WiFi hotspot as well.

Schedule

I’m still a bit behind schedule, but because I was able to connect the IP cameras and the Raspberry Pi to the same network a good portion of the camera integration has been taken care of. Once I get the footage from the real-life intersection I will train the vehicle detection Haar classifier, the code for which I already have typed out. Also, I no longer feel that it’s necessary to use the BLE cameras now that the IP cameras and Raspberry Pi have connected to the same network. If I have trouble accessing the IP camera video stream on the Raspberry Pi, I will revisit that as an option.

Deliverables

By the end of next week, I will:

• Train the vehicle detection Haar classifier
• Access the IP camera video feed from the Raspberry Pi and write code that will allow me to access that feed frame by frame to run the object detection model on it

Work Done

This week, I helped out with the proposal presentation slides and did some implementation planning and parts research, particularly for the camera setup. In particular, I made the solution approach and testing, verification, and metrics slides (with input from my team members to make sure we were all on the same page.) Below is the block diagram I developed for our system.

For the rest of this week, I’ve been looking into how we’re going to connect the IP Cameras (which will ideally be at each intersection) to the RPi. CMU’s WiFi is notoriously difficult to connect to with external devices (like an RPi – but that’s been done before. It’ll be a particular pain – if even possible – to connect the IP cameras), but we have a few options. We can potentially connect the RPi to CMU WiFi and use it as a hotspot of sorts that the cameras can then connect to (however, this would probably require us to use an intersection other than Fifth and Craig to test our system.) We can also purchase a mobile hotspot and connect the RPi and cameras to that.

I also looked into BLE camera setups and couldn’t find any substantive projects that were similar in scope to ours. These cameras cannot stream video, and furthermore require wired connections to BLE microcontrollers like the Arduino Nano 33 BLE or ESP32. If we can’t get the IP camera setup to work (the idea is to order one IP camera to start with to see how the setup goes), we will probably default to wired cameras (standard Raspberry Pi cameras) for one or two sides of the intersection and simulate the other sides for demo purposes.

Schedule

Progress is mostly on schedule — I did quite a bit of research on the different kinds of cameras we can use and thought about different ways of putting our whole system together. Before deciding on a camera implementation, however, I want to meet with Prof. Sullivan and Mukundh and discuss the feasibility of what we have in mind. In that sense I am slightly behind schedule as I was supposed to finish up camera research by Monday.

Tasks this Week

• Decide on a camera implementation and get some ordered so we can start setting things up.
• Get my hands on a Raspberry Pi 4 (hopefully there are some in the course inventory) and boot it up, then see if I can host a hotspot on the Pi itself for other devices to connect to.
• Start writing the object detection algorithm for cars and pedestrians and test it on old traffic camera footage.