Early Intent Prediction of Vulnerable Road Users from Visual Attributes Using Multi-Task Learning Network. Saleh, K., Hossny, M., & Nahavandi, S. In 2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pages 3367–3372, October, 2017. doi abstract bibtex In this paper we are presenting a novel approach for the problem of vulnerable road users (VRUs) attribute prediction which play such critical role for the intent prediction models of VRUs. We formulated the problem as a multi-task learning (MTL) image classification problem and we utilized a convolution neural network (ConvNet) based technique to exploit the commonality between two of the most important attributes of VRUs for intent prediction models (i.e, head orientation and body posture). We achieved classification accuracy scores of 83% and 76% for the body posture and head orientation attributes respectively. We compared the performance of our proposed solution against individual single task learning ConvNet models for each attribute and achieved significant overall accuracy over the two attribute classification tasks. Furthermore, we compared our proposed MTL-ConvNet model against other MTL approaches and achieved more than 18% AP score improvement in the classification of body posture attribute.
@inproceedings{saleh2017early,
title = {Early Intent Prediction of Vulnerable Road Users from Visual Attributes Using Multi-Task Learning Network},
booktitle = {2017 {{IEEE International Conference}} on {{Systems}}, {{Man}}, and {{Cybernetics}} ({{SMC}})},
author = {Saleh, Khaled and Hossny, Mohammed and Nahavandi, Saeid},
year = {2017},
month = oct,
pages = {3367--3372},
doi = {10.1109/SMC.2017.8123150},
urldate = {2024-03-08},
abstract = {In this paper we are presenting a novel approach for the problem of vulnerable road users (VRUs) attribute prediction which play such critical role for the intent prediction models of VRUs. We formulated the problem as a multi-task learning (MTL) image classification problem and we utilized a convolution neural network (ConvNet) based technique to exploit the commonality between two of the most important attributes of VRUs for intent prediction models (i.e, head orientation and body posture). We achieved classification accuracy scores of 83\% and 76\% for the body posture and head orientation attributes respectively. We compared the performance of our proposed solution against individual single task learning ConvNet models for each attribute and achieved significant overall accuracy over the two attribute classification tasks. Furthermore, we compared our proposed MTL-ConvNet model against other MTL approaches and achieved more than 18\% AP score improvement in the classification of body posture attribute.},
annotation = {17 citations (Semantic Scholar/DOI) [2024-04-26]},
file = {C\:\\Users\\gregf\\Zotero\\storage\\2SLZYRAW\\Saleh et al. - 2017 - Early intent prediction of vulnerable road users f.pdf;C\:\\Users\\gregf\\Zotero\\storage\\8UERIV8K\\8123150.html}
}
Downloads: 0
{"_id":"JqynxbxiEoSQpdwh6","bibbaseid":"saleh-hossny-nahavandi-earlyintentpredictionofvulnerableroadusersfromvisualattributesusingmultitasklearningnetwork-2017","author_short":["Saleh, K.","Hossny, M.","Nahavandi, S."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","title":"Early Intent Prediction of Vulnerable Road Users from Visual Attributes Using Multi-Task Learning Network","booktitle":"2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC)","author":[{"propositions":[],"lastnames":["Saleh"],"firstnames":["Khaled"],"suffixes":[]},{"propositions":[],"lastnames":["Hossny"],"firstnames":["Mohammed"],"suffixes":[]},{"propositions":[],"lastnames":["Nahavandi"],"firstnames":["Saeid"],"suffixes":[]}],"year":"2017","month":"October","pages":"3367–3372","doi":"10.1109/SMC.2017.8123150","urldate":"2024-03-08","abstract":"In this paper we are presenting a novel approach for the problem of vulnerable road users (VRUs) attribute prediction which play such critical role for the intent prediction models of VRUs. We formulated the problem as a multi-task learning (MTL) image classification problem and we utilized a convolution neural network (ConvNet) based technique to exploit the commonality between two of the most important attributes of VRUs for intent prediction models (i.e, head orientation and body posture). We achieved classification accuracy scores of 83% and 76% for the body posture and head orientation attributes respectively. We compared the performance of our proposed solution against individual single task learning ConvNet models for each attribute and achieved significant overall accuracy over the two attribute classification tasks. Furthermore, we compared our proposed MTL-ConvNet model against other MTL approaches and achieved more than 18% AP score improvement in the classification of body posture attribute.","annotation":"17 citations (Semantic Scholar/DOI) [2024-04-26]","file":"C\\:\\\\Users\\\\gregf\\\\Zotero\\\\storage\\\\2SLZYRAW\\§aleh et al. - 2017 - Early intent prediction of vulnerable road users f.pdf;C\\:\\\\Users\\\\gregf\\\\Zotero\\\\storage\\\\8UERIV8K\\\\8123150.html","bibtex":"@inproceedings{saleh2017early,\n title = {Early Intent Prediction of Vulnerable Road Users from Visual Attributes Using Multi-Task Learning Network},\n booktitle = {2017 {{IEEE International Conference}} on {{Systems}}, {{Man}}, and {{Cybernetics}} ({{SMC}})},\n author = {Saleh, Khaled and Hossny, Mohammed and Nahavandi, Saeid},\n year = {2017},\n month = oct,\n pages = {3367--3372},\n doi = {10.1109/SMC.2017.8123150},\n urldate = {2024-03-08},\n abstract = {In this paper we are presenting a novel approach for the problem of vulnerable road users (VRUs) attribute prediction which play such critical role for the intent prediction models of VRUs. We formulated the problem as a multi-task learning (MTL) image classification problem and we utilized a convolution neural network (ConvNet) based technique to exploit the commonality between two of the most important attributes of VRUs for intent prediction models (i.e, head orientation and body posture). We achieved classification accuracy scores of 83\\% and 76\\% for the body posture and head orientation attributes respectively. We compared the performance of our proposed solution against individual single task learning ConvNet models for each attribute and achieved significant overall accuracy over the two attribute classification tasks. Furthermore, we compared our proposed MTL-ConvNet model against other MTL approaches and achieved more than 18\\% AP score improvement in the classification of body posture attribute.},\n annotation = {17 citations (Semantic Scholar/DOI) [2024-04-26]},\n file = {C\\:\\\\Users\\\\gregf\\\\Zotero\\\\storage\\\\2SLZYRAW\\\\Saleh et al. - 2017 - Early intent prediction of vulnerable road users f.pdf;C\\:\\\\Users\\\\gregf\\\\Zotero\\\\storage\\\\8UERIV8K\\\\8123150.html}\n}\n\n","author_short":["Saleh, K.","Hossny, M.","Nahavandi, S."],"key":"saleh2017early","id":"saleh2017early","bibbaseid":"saleh-hossny-nahavandi-earlyintentpredictionofvulnerableroadusersfromvisualattributesusingmultitasklearningnetwork-2017","role":"author","urls":{},"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://bibbase.org/network/files/aq4q4TBa82qnNuJ2R","dataSources":["PhJMrkQA7r5Go9zar","4NxBSuKtJxKDWdWso","Mcex2YsbPdMjaPz4T"],"keywords":[],"search_terms":["early","intent","prediction","vulnerable","road","users","visual","attributes","using","multi","task","learning","network","saleh","hossny","nahavandi"],"title":"Early Intent Prediction of Vulnerable Road Users from Visual Attributes Using Multi-Task Learning Network","year":2017}