The object and event-level video annotation tool for anomaly detection
There are many quality datasets that are being used in anomalous detection systems. However, with the range of possible machine learning approaches, each of which has specific input annotation requirements e.g. levels of annotation whether frame level or object level, types of input whether image or...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | text |
| 語言: | English |
| 出版: |
Animo Repository
2022
|
| 主題: | |
| 在線閱讀: | https://animorepository.dlsu.edu.ph/etdb_comtech/6 https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1003&context=etdb_comtech |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | De La Salle University |
| 語言: | English |
| 總結: | There are many quality datasets that are being used in anomalous detection systems. However, with the range of possible machine learning approaches, each of which has specific input annotation requirements e.g. levels of annotation whether frame level or object level, types of input whether image or video, types of annotation whether fixation data or object location, types of anomalous events etc., the development of the anomalous event detection system is limited by the type available annotated dataset. Although video annotation tools exist to fill these gaps, they are mostly tailored towards object detection systems. This study aims to develop an annotation tool specifically for labeling anomalous road events that can aid in creating datasets for training machine learning models. The proposed system patterned on existing video annotation systems which are used to annotate objects and events inside a video. There are two features that differentiate it from non-anomaly annotation systems. The first feature is a dedicated post processing module, which is used to package the final output towards anomaly detection. The second feature is an automation module which is used to automate the annotation process specifically for annotating anomalous events. |
|---|