DoTA: Unsupervised Detection of Traffic Anomaly in Driving Videos

Yao, Yu; Wang, Xizi; Xu, Mingze; Pu, Zelin; Wang, Yuchen; Atkins, Ella M.; Crandall, David J.

doi:10.1109/tpami.2022.3150763

Cited by 37 publications

(33 citation statements)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While straightforward, it is a highly complex problem to solve in real-world applications because there could be many known and unknown casualties. Very recently, Yao et al [17] proposed to localize risky objects by predicting future trajectory of traffic participants over a short horizon. Inconsistent predictions indicate that an anomaly has occurred or is about to happen.…”

Section: Risky Object Localization In a Driving Scenementioning

confidence: 99%

“…These datasets are not directly applicable to the problem of risky object localization due to the lack of objectlevel risk annotation. Recently, Yao et al [17] collected 4,600 videos with the risky object annotation. They annotated the risky objects contributing to an accident with their bounding boxes in videos when annotators judge the accident seems inevitable.…”

Section: Risky Object Localization Datasetsmentioning

confidence: 99%

“…Since early detection of accidentinvolved objects is essential, this study annotates the time when a risky object appears in the driving scene for the first time, termed risky object appearance time. Unlike [17] that annotates the anomaly beginning time as the moment when annotators think an accident is inevitable, the approach to annotating risky object appearance time in this paper does not need annotators' subjective judgment. The accident beginning time is when a vehicle collides with another static or dynamic object.…”

Section: B Data Annotationmentioning

confidence: 99%

“…This study compares the AM-Net (both without and with the attention module) trained on ROL (i.e., models #8 and #10 in TABLE III) to existing models [16], [17], [19], [20] on the testing dataset of DoTA [17] from the perspective of video anomaly detection. That is, they are compared on the metric of frame-level AUC.…”

Section: Comparison To the State-of-the-art Modelsmentioning

confidence: 99%

“…Lastly, the long-term temporal dependency of traffic agents underlying the accident risk is hard to capture. Several attempts have explored this topic, mainly using deep learning methods [12]- [17]. While these studies have laid a solid foundation for risky object localization, the opportunity to further improve the performance by extracting rich information from RGB videos is worth an exploration, especially when the training dataset is limited.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

An Attention-guided Multistream Feature Fusion Network for Localization of Risky Objects in Driving Videos

Karim¹,

Qin²,

Yin³

2022

Preprint

View full text Add to dashboard Cite

Detecting dangerous traffic agents in videos captured by vehicle-mounted dashboard cameras (dashcams) is essential to facilitate safe navigation in a complex environment. Accident-related videos are just a minor portion of the driving video big data, and the transient pre-accident processes are highly dynamic and complex. Besides, risky and non-risky traffic agents can be similar in their appearance. These make risky object localization in the driving video particularly challenging. To this end, this paper proposes an attention-guided multistream feature fusion network (AM-Net) to localize dangerous traffic agents from dashcam videos. Two Gated Recurrent Unit (GRU) networks use object bounding box and optical flow features extracted from consecutive video frames to capture spatiotemporal cues for distinguishing dangerous traffic agents. An attention module coupled with the GRUs learns to attend to the traffic agents relevant to an accident. Fusing the two streams of features, AM-Net predicts the riskiness scores of traffic agents in the video. In supporting this study, the paper also introduces a benchmark dataset called Risky Object Localization (ROL). The dataset contains spatial, temporal, and categorical annotations with the accident, object, and scene-level attributes. The proposed AM-Net achieves a promising performance of 85.73% AUC on the ROL dataset. Meanwhile, the AM-Net outperforms current state-of-the-art for video anomaly detection by 6.3% AUC on the DoTA dataset. A thorough ablation study further reveals AM-Net's merits by evaluating the contributions of its different components.

show abstract

Section: Risky Object Localization In a Driving Scenementioning

confidence: 99%

Section: Risky Object Localization Datasetsmentioning

confidence: 99%

Section: B Data Annotationmentioning

confidence: 99%

Section: Comparison To the State-of-the-art Modelsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

An Attention-guided Multistream Feature Fusion Network for Localization of Risky Objects in Driving Videos

Karim¹,

Qin²,

Yin³

2022

Preprint

View full text Add to dashboard Cite

show abstract

Artificialintelligence‐basedremote monitoring system for automated anomaly detection in wireless sensor networks

Govardhan¹,

Surendranath

Sundararajan³

2022

Concurrency and Computation

View full text Add to dashboard Cite

In recent decades, wireless sensor networks (WSNs) are employed in the remote monitoring system which experiences a huge delay due to the excessive time consumption by the image processing techniques. The existing video surveillance system is unable to detect the anomaly accurately and falls short in classifying the routine activity and the anomaly activities. Therefore, an efficient anomaly detection model is necessitated for identifying the anomaly behavior of malicious nodes in the WSN environment. This research work introduces a novel artificial intelligence (AI)-based remote monitoring system for WSN. Primarily, the online video frames are retrieved from the wireless video sensors where the frames are generated by segmenting the entire stream of video as a constant-sized structural frame. The sequence/stream of video segments is fed into the proposed AI framework to detect anomaly activities. It has multiple levels of hidden layers to compare the features of the input captured image with the preloaded reference image. Furthermore, a 5-fold cross-validation approach has been inspected to validate the efficacy of the suggested AI model. The performance results evident that the proposed AI system acquires a better prediction accuracy of 98.2% and precision of 97.3% than the existing monitoring systems.

show abstract

Accident-CLIP: Text-Video Benchmarking for Fine-Grained Accident Classification in Driving Scenes

Yang,

Fang,

Zhu

et al. 2024

Lecture Notes in Electrical Engineering

View full text Add to dashboard Cite

DoTA: Unsupervised Detection of Traffic Anomaly in Driving Videos

Cited by 37 publications

References 66 publications

An Attention-guided Multistream Feature Fusion Network for Localization of Risky Objects in Driving Videos

An Attention-guided Multistream Feature Fusion Network for Localization of Risky Objects in Driving Videos

Artificialintelligence‐basedremote monitoring system for automated anomaly detection in wireless sensor networks

Accident-CLIP: Text-Video Benchmarking for Fine-Grained Accident Classification in Driving Scenes

Contact Info

Product

Resources

About