PedestrianDetect

• 

  Algorithm detects my body, and people behind me.

• 

  Memory usage of program in mebibytes.

• 

  System detects people sitting in the background.

• 

  Areas of high pedestrian concentration are marked. This includes me and those to my right.

Inspiration

I had a vision of making vehicles safer for people by using machine learning to assist drivers in recognizing pedestrians at night.

What it does

Detects people and parts of people in live video frames using bounding boxes and sounds an alert once detected.

How I built it

I built the system by training weights using the yolo algorithm on images of people. Next, I wrote OpenCV code to add the clear bounding boxes to mark people in video frames. I also adjusted the threshold value to accurately detect people and minimize the chance of picking up noise in video feeds. Finally, I added some code to trace the memory usage of the program.

Challenges I ran into

It was difficult to successfully lower the opacity of the bounding boxes while maintaining the quality of the video frames.

Accomplishments that I'm proud of

Completing a semi-supervised machine learning project that may potentially save lives.

What I learned

Semi-supervised learning requires a moderate amount of per-processing, but the final result is worth the work.

What's next for PedestrianDetect

I want to take a deep learning approach in order to detect people more accurately. I also plan on testing the system in a closed course setting to improve the technology's road deployment potential.

Built With

• opencv
• python
• yolo