https://github.com/yannickfunk/streestsignrecognition
Street Sign Recognition for Autonomous Driving
https://github.com/yannickfunk/streestsignrecognition
Last synced: 16 days ago
JSON representation
Street Sign Recognition for Autonomous Driving
- Host: GitHub
- URL: https://github.com/yannickfunk/streestsignrecognition
- Owner: yannickfunk
- Created: 2019-09-29T22:55:53.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2019-09-29T22:59:32.000Z (over 6 years ago)
- Last Synced: 2025-03-06T01:47:36.177Z (over 1 year ago)
- Language: Python
- Size: 16.6 KB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Traffic Sign Recognition
#### Description
In this project I built a 97% accurate classifier for street sign recognition. I used the large German Traffic Sign Recognition Benchmark data set https://www.kaggle.com/meowmeowmeowmeowmeow/gtsrb-german-traffic-sign which contains over 40 classes for training a convolutional neural network.
#### Training Results
After training 800 epochs using 30% of the data for validation, I obtained a validation accuracy of 97%.
The following confusion matrix shows the predictions of the network:
As seen, the classifier is very robust, due to nearly zero entries on the side diagonals.
The classification report:
precision recall f1-score support
0 1.00 1.00 1.00 84
1 0.98 0.96 0.97 888
2 1.00 1.00 1.00 804
3 0.98 0.99 0.99 528
4 0.98 1.00 0.99 840
5 1.00 1.00 1.00 864
6 0.97 1.00 0.99 312
7 0.96 0.96 0.96 252
8 0.99 0.99 0.99 168
9 1.00 1.00 1.00 444
10 0.96 1.00 0.98 480
11 1.00 0.94 0.97 84
12 0.99 0.97 0.98 900
13 0.91 0.96 0.93 144
14 0.96 0.99 0.98 132
15 0.97 1.00 0.98 156
16 0.99 0.93 0.96 204
17 0.96 0.98 0.97 108
18 0.99 0.95 0.97 600
19 0.99 0.93 0.96 240
20 0.99 1.00 0.99 96
21 0.93 1.00 0.96 216
22 0.76 0.74 0.75 108
23 0.99 1.00 0.99 564
24 1.00 0.87 0.93 180
25 0.98 1.00 0.99 312
26 1.00 0.69 0.81 96
27 0.88 0.92 0.90 275
28 0.80 0.81 0.80 168
29 0.99 1.00 0.99 480
30 0.95 0.88 0.92 156
31 0.79 0.65 0.71 84
32 0.91 0.96 0.93 828
33 0.80 0.58 0.67 120
34 0.99 0.99 0.99 792
35 0.82 0.97 0.89 144
36 1.00 1.00 1.00 96
37 0.99 1.00 0.99 96
38 0.97 0.97 0.97 744
39 1.00 0.98 0.99 168
40 0.96 0.97 0.96 576
41 0.96 0.99 0.98 564
42 0.99 1.00 1.00 588
accuracy 0.97 15683
macro avg 0.95 0.94 0.95 15683
weighted avg 0.97 0.97 0.97 15683
shows that there are some classes, not recognized well (e.g 31), but this correlates directly with the lack of training data supporting this class. Except for the underrepresented classes, the prediction values are very good.