In this work, we aim to detect car crash accidents in video. We propose a three-stage framework: The first one is a car detection method using convolutional neural networks, in this case, we used the net You Only Look Once (YOLO); the second stage is a tracker in order to focus each car; then the final stage for each car we use the Violent Flow (ViF) descriptor with a Support Vector Machine (SVM) in order to detect the car crashes. Our proposal is almost in real time with just 0.5 seconds of delay and also we got a 89% accuracy detecting car crashes.
|Title of host publication||Proceedings - 2018 44th Latin American Computing Conference, CLEI 2018|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||6|
|State||Published - Oct 2018|
|Event||44th Latin American Computing Conference, CLEI 2018 - Sao Paulo, Brazil|
Duration: 1 Oct 2018 → 5 Oct 2018
|Name||Proceedings - 2018 44th Latin American Computing Conference, CLEI 2018|
|Conference||44th Latin American Computing Conference, CLEI 2018|
|Period||1/10/18 → 5/10/18|
Bibliographical notePublisher Copyright:
© 2018 IEEE.
- Car Crash
- Convolutional Networks
- Deep Learning