The main objective of this project is to identify overspeed vehicles, using Deep Learning and Machine Learning Algorithms. After acquisition of series of images from the video, trucks are detected using Haar Cascade Classifier. The model for the classifier is trained using lots of positive and negative images to make an XML file. This is followed by tracking down the vehicles and estimating their speeds with the help of their respective locations, ppm (pixels per meter) and fps (frames per second). Now, the cropped images of the identified trucks are sent for License Plate detection. The CCA (Connected Component Analysis) assists in Number Plate detection and Characters Segmentation. The SVC model is trained using characters images (20X20) and to increase the accuracy, 4 cross fold validation (Machine Learning) is also done. This model aids in recognizing the segmented characters. After recognition, the calculated speed of the trucks is fed into an excel sheet along with their license plate numbers. These trucks are also assigned some IDs to generate a systematized database.
158
stars
38
forks
source link
Enhancements for Speed Estimation and License Plate Detection Project #12
I've recently had the opportunity to review your fascinating project on vehicle speed estimation and license plate detection. It's a commendable effort in utilising machine learning and deep learning algorithms for identifying overspeed vehicles. However, I'd like to suggest a few enhancements and address some concerns.
Integration of Modern Object Detection Techniques: The project currently uses Haar Cascade Classifier for truck detection and CCA for number plate detection. While these are robust methods, they might be surpassed by more contemporary techniques. Incorporating state-of-the-art object detection models like YOLOv3 could significantly improve accuracy and efficiency.
Improving Speed Estimation Accuracy: The current methodology for speed estimation seems to rely heavily on the frames per second (fps) and pixels per meter (ppm) calculations, which might not always yield precise results. Could we consider integrating additional sensors or GPS data to enhance accuracy?
Handling Environmental Variabilities: The project acknowledges limitations due to environmental factors like dust on license plates or varying scales of the object. Are there plans to implement preprocessing steps or additional algorithms to mitigate these issues?
Training on Diverse Vehicle Models: The limitation of the system's ability to identify only newer truck models could be a significant hindrance. Expanding the training dataset to include a wider range of vehicle models, including older ones, might enhance the system's applicability.
Optimising Processing Time: The project mentions the expensive scanning and processing time as a limitation. Are there any plans to optimise the code or utilise more efficient algorithms to reduce this?
License Plate Recognition in Challenging Conditions: Lastly, the difficulty in detecting license plates under certain conditions like being obscured or dirty. Are there any strategies in place to improve recognition in such scenarios?
I'm looking forward to your thoughts on these suggestions. Your project has immense potential, and I believe these enhancements could make it even more impactful.
Dear Team,
I've recently had the opportunity to review your fascinating project on vehicle speed estimation and license plate detection. It's a commendable effort in utilising machine learning and deep learning algorithms for identifying overspeed vehicles. However, I'd like to suggest a few enhancements and address some concerns.
Integration of Modern Object Detection Techniques: The project currently uses Haar Cascade Classifier for truck detection and CCA for number plate detection. While these are robust methods, they might be surpassed by more contemporary techniques. Incorporating state-of-the-art object detection models like YOLOv3 could significantly improve accuracy and efficiency.
Improving Speed Estimation Accuracy: The current methodology for speed estimation seems to rely heavily on the frames per second (fps) and pixels per meter (ppm) calculations, which might not always yield precise results. Could we consider integrating additional sensors or GPS data to enhance accuracy?
Handling Environmental Variabilities: The project acknowledges limitations due to environmental factors like dust on license plates or varying scales of the object. Are there plans to implement preprocessing steps or additional algorithms to mitigate these issues?
Training on Diverse Vehicle Models: The limitation of the system's ability to identify only newer truck models could be a significant hindrance. Expanding the training dataset to include a wider range of vehicle models, including older ones, might enhance the system's applicability.
Optimising Processing Time: The project mentions the expensive scanning and processing time as a limitation. Are there any plans to optimise the code or utilise more efficient algorithms to reduce this?
License Plate Recognition in Challenging Conditions: Lastly, the difficulty in detecting license plates under certain conditions like being obscured or dirty. Are there any strategies in place to improve recognition in such scenarios?
I'm looking forward to your thoughts on these suggestions. Your project has immense potential, and I believe these enhancements could make it even more impactful.
Best regards, yihong1120