MACHINE LEARNING MODEL FOR REAL-TIME MOTORCYCLE VIOLATION DETECTION
Traffic violations are still a common thing both by car drivers and motorcycle. Traffic violators are usually supervised by traffic police. In Indonesia, traffic supervision is still processed manually by the traffic police. This research tries to develop a machine learning model that can automatica...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/43655 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Traffic violations are still a common thing both by car drivers and motorcycle. Traffic violators are usually supervised by traffic police. In Indonesia, traffic supervision is still processed manually by the traffic police. This research tries to develop a machine learning model that can automatically detect violations by motorcyclist in real-time performance. The approach used in this study is to adopt object detection algorithm YOLOv3 (You Only Look Once) to detect related objects in the case of violation detection, these objects are the helmet, people, and motorcycle. The dataset used in this study is a traffic video dataset collected through recording with a CCTV camera at 3 meters height. The violation object detection approach from this study obtained a mAP (mean average precision) score of 0,935 (helmet), 0,923 (person), and 0,970 (motorcycle). The model run at speeds of 51,24 frames per second (real-time) on NVIDIA GTX 1080 Ti device. The object detection model then adapted for the purpose of violations detection, the model obtained an accuracy of 90.1 % for detecting riders without a helmet and 33,3 % for detecting excess passengers violation (more than 2 persons in one motorcycle). This study is expected to be a catalyst for auto surveillance/violation detection system in Indonesia. |
|---|