Car over-speeding detection using time-distance approximation
Existing speed traps use cameras installed at fixed positions to detect the overspeeding vehicle. However, observant drivers can evasively slow down when approaching the proximity of the speed cameras. This project proposes Catch, a less evasible overspeeding detection using time-to-distance approxi...
保存先:
| 第一著者: | |
|---|---|
| フォーマット: | Final Year Project / Dissertation / Thesis |
| 出版事項: |
2022
|
| 主題: | |
| オンライン･アクセス: | http://eprints.utar.edu.my/4655/1/fyp_CS_2022_LYL.pdf http://eprints.utar.edu.my/4655/ |
| タグ: |
タグ追加
タグなし, このレコードへの初めてのタグを付けませんか!
|
| id |
my-utar-eprints.4655 |
|---|---|
| record_format |
eprints |
| spelling |
my-utar-eprints.46552022-10-13T07:27:20Z Car over-speeding detection using time-distance approximation Lai, Yu Liang Q Science (General) T Technology (General) Existing speed traps use cameras installed at fixed positions to detect the overspeeding vehicle. However, observant drivers can evasively slow down when approaching the proximity of the speed cameras. This project proposes Catch, a less evasible overspeeding detection using time-to-distance approximation. Instead of using on-the-spot speed data, Catch calculates the average driving time for a journey (point A to B) and compare the elapsed time against the permitted timestamps based on the route-specific speed limit. For example, a vehicle is flagged as overspeeding if it completes a 100KM journey within 1 hour on a route with a speed limit of 100KM/h. Catch use machine learning to first locate the car plate in an image using a car plate localization model trained with open-source framework TensorFlow Object Detection API, and second to identify vehicles using a car plate recognition model trained with a CRNN. The car plate recognition model is trained using local car plates images that includes ‘A’, ‘P’ and ‘W’ car plates. We leverage on an existing CRNN model for transfer learning; we discard the fully-connected layers at the end of CNN, squeeze the CNN before connecting it to RNN. The resulting model is highly accurate, scoring 0.002894 val_loss and 95.68% accuracy in predicting unseen Malaysian Car plates. Then, we set up two cameras to run on Jetson Nano for real-time passing car footage at two checkpoints. The same car plates matched from two checkpoints will be calculated for average speed using time difference between two checkpoints for overspeeding detection. Catch is able to catch 1853 overspeeding cars that had been escaped from the traditional speed trap in a month. 2022-04-22 Final Year Project / Dissertation / Thesis NonPeerReviewed application/pdf http://eprints.utar.edu.my/4655/1/fyp_CS_2022_LYL.pdf Lai, Yu Liang (2022) Car over-speeding detection using time-distance approximation. Final Year Project, UTAR. http://eprints.utar.edu.my/4655/ |
| institution |
Universiti Tunku Abdul Rahman |
| building |
UTAR Library |
| collection |
Institutional Repository |
| continent |
Asia |
| country |
Malaysia |
| content_provider |
Universiti Tunku Abdul Rahman |
| content_source |
UTAR Institutional Repository |
| url_provider |
http://eprints.utar.edu.my |
| topic |
Q Science (General) T Technology (General) |
| spellingShingle |
Q Science (General) T Technology (General) Lai, Yu Liang Car over-speeding detection using time-distance approximation |
| description |
Existing speed traps use cameras installed at fixed positions to detect the overspeeding vehicle. However, observant drivers can evasively slow down when approaching the proximity of the speed cameras. This project proposes Catch, a less evasible overspeeding detection using time-to-distance approximation. Instead of using on-the-spot speed data, Catch calculates the average driving time for a journey (point A to B) and compare the elapsed time against the permitted timestamps based on the route-specific speed limit. For example, a vehicle is flagged as overspeeding if it completes a 100KM journey within 1 hour on a route with a speed limit of 100KM/h. Catch use machine learning to first locate the car plate in an image using a car plate localization model trained with open-source framework TensorFlow Object Detection API, and second to identify vehicles using a car plate recognition model trained with a CRNN. The car plate recognition model is trained using local car plates images that includes ‘A’, ‘P’ and ‘W’ car plates. We leverage on an existing CRNN model for transfer learning; we discard the fully-connected layers at the end of CNN, squeeze the CNN before connecting it to RNN. The resulting model is highly accurate, scoring 0.002894 val_loss and 95.68% accuracy in predicting unseen Malaysian Car plates. Then, we set up two cameras to run on Jetson Nano for real-time passing car footage at two checkpoints. The same car plates matched from two checkpoints will be calculated for average speed using time difference between two checkpoints for overspeeding detection. Catch is able to catch 1853 overspeeding cars that had been escaped from the traditional speed trap in a month. |
| format |
Final Year Project / Dissertation / Thesis |
| author |
Lai, Yu Liang |
| author_facet |
Lai, Yu Liang |
| author_sort |
Lai, Yu Liang |
| title |
Car over-speeding detection using time-distance approximation
|
| title_short |
Car over-speeding detection using time-distance approximation
|
| title_full |
Car over-speeding detection using time-distance approximation
|
| title_fullStr |
Car over-speeding detection using time-distance approximation
|
| title_full_unstemmed |
Car over-speeding detection using time-distance approximation
|
| title_sort |
car over-speeding detection using time-distance approximation |
| publishDate |
2022 |
| url |
http://eprints.utar.edu.my/4655/1/fyp_CS_2022_LYL.pdf http://eprints.utar.edu.my/4655/ |
| _version_ |
1748184794152304640 |
| score |
13.250246 |