- Virtual Env and OpenCV Look into the steps mentioned inside OpenCV Folder
- Remaining python libraries for Car Detection in the same virtual env where you have installed OpenCV
Install the libraries with correct versions mentioned in requirements.txt, use the command like
pip install -r requirements.txt
Get Inside the python virtual environment created during installation process.
Run juypter notebook and open vechile detection.ipynb
Run this Python Notebook for simulating the results.
The following are some of our results.
- The weights obtained after training the model are tiny-yolo.weights (181 Mb). They are to be downloaded and used.
- For the images we would have to resize them , one can use command like
mogrify -resize 1280x720 *.jpgto in place resize the images Reference
* Training the model further for Bike, Cab and partial Image detection.
* Developing the network graph of traffic data, eg. Like simple count of total vechiles on any signal, taking them as weights on that node(traffic signal) with roads as edges connecting to another node(traffic signal).
* We shall be then solving network optimization problem at any given instance for calculating the most optimal timings for the signals.
* More features like speed detection, number plate detection, helmet & seat beat detections, etc, can be further added.
- We were selected within Top 35 teams with this project.
- Algorithm performance was poor as we were not able to bring out the total efficiency of the YOLO model. More fine tuning, understanding of the model, keras and above alldedicated effort is required.
- This project will be on hold till my schedule and time permits.