Story Behind the Project
I complained to my good friends, who can tolerate my late night bombardment, about how I needed a project to keep my mind running productively and busy, aside from exercise and actually working on my real tasks that are much more rewarding later on (e.g. practice technical interviews, clean up my portfolio). I wanted work, where I could look forward to every return home.
My friend said that he'd be ready to test my app, so I guess I'll have to create the app, and do my best to stick to that.
I surely want to test out my app. Although I am not a fan of driving, I've been in the back seat long enough to know that not all traffic lights are perfect and can be a dreadful patience and discipline tester in the same spot for 5+ minutes. This includes sitting in traffic for quite, where we might as well nap there, too.
That's enough to catch at least 25% of a power nap.
Every stopping moment, the app starts finding traffic lights and detects another car movement to check whether you should press the gas pedal or not. The app should also allow you to exit and still run in the background so you can do simple tasks such as check e-mails and playing music. And, shouldn't use so much battery life.
Car Movement Detection
As much as I want to train my own haar cascade, I am impatient and want a running prototype (also this is not even paid and out of learning interest) so I'm going to copy this project: https://www.geeksforgeeks.org/opencv-python-program-vehicle-detection-video-frame/
Or, I can do HSV segmentation: https://www.researchgate.net/publication/251423714_Vehicle_Detection_at_Night_Based_on_TailLight_Detection
American Traffic Lights
There are light patterns for American traffic lights. The light sequence is red, then green to go, and green, flashing yellow, and then red to stop. And, it's vertical rows of 3 circles.
Also, this is a good read: https://autottblog.wordpress.com/programming-the-car/opencv/
Traffic Light and Color Detection
Find 3 circles that are red, yellow, and green in a vertical row, and calculate the distance between those circles to determine if it is a traffic light. By converting the image to HSV to find the 3 circles and its distance.
Pseudocode and Algorithm for Traffic Light Detection
Thanks, researchers, for making it easy for me: https://www.researchgate.net/publication/300581967_Traffic_Light_Recognition_System_for_People_with_Color_Blindness/download
Basically, segment frames into HSV then LOTS AND LOTS of tweaking the threshold. They made a nice flowchart in there. Thank goodness.
No comments:
Post a Comment