Pose Detection of Dead Body in Crime Scene Investigation
Issued Date
2023-10-27
Resource Type
Scopus ID
2-s2.0-85187555202
Journal Title
ACM International Conference Proceeding Series
Start Page
168
End Page
173
Rights Holder(s)
SCOPUS
Bibliographic Citation
ACM International Conference Proceeding Series (2023) , 168-173
Suggested Citation
Werukanjana P., Permpool N., Sa-Nga-Ngam P., Muttitanon W. Pose Detection of Dead Body in Crime Scene Investigation. ACM International Conference Proceeding Series (2023) , 168-173. 173. doi:10.1145/3633637.3633662 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/97659
Title
Pose Detection of Dead Body in Crime Scene Investigation
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
This paper explores the potential of real-time object detection in Crime Scene Investigation to assist investigators in determining the cause of death and bringing justice to those who deserve it. The study discusses the critical steps in implementing dead body pose estimation using the custom dataset with YOLOv8, including data collection, model training, fine-tuning, and testing results. However, human expertise and integration are necessary in the subsequent stages of crime scene investigation to enhance the effectiveness of the proposed system. The experiment uses Automatic Mixed Precision (AMP) by bbox and poses estimation, with suitable performance matrices of the bbox. The result shows all class mPA50 at 0.98 of bbox prediction, while the mPA50 of the pose estimation is 0.969. When dealing with small custom datasets, the precision needed for predicting keypoints results in lower performance of mAP values compared to bbox detection.