Towards Real-Time Systems for Vehicle Re-Identification, Multi-Camera Tracking, and Anomaly Detection

Peri, Neehar; Khorramshahi, Pirazh; Rambhatla, Sai Saketh; Shenoy, Vineet; Rawat, Saumya; Chen, Jun-Cheng; Chellappa, Rama

doi:10.1109/cvprw50498.2020.00319

Cited by 43 publications

(14 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Vehicle tracking is one of the most important tasks in MOT owing to its extensive applications, such as traffic monitoring, analysis, and control [ 8 ]. The vehicle-tracking task has several challenges, including tracking objects in various sizes and views, in different environments of illuminations, and with heavy occlusions.…”

Section: Related Workmentioning

confidence: 99%

“…In recent years, numerous studies have been conducted to implement multi-object tracking (MOT), which is a process that predicts trajectories of multiple objects detected across a sequence of video frames [ 1 , 2 , 3 , 4 , 5 , 6 , 7 ]. Among all the approaches, deep learning-based methods have shown remarkable improvements and have enabled several applications, such as video surveillance and analysis [ 8 ], human recognition and communication [ 9 ], and autonomous driving [ 10 ].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Efficient Single-Shot Multi-Object Tracking for Vehicles in Traffic Scenarios

Lee

Yoo

et al. 2021

Sensors

View full text Add to dashboard Cite

Multi-object tracking is a significant field in computer vision since it provides essential information for video surveillance and analysis. Several different deep learning-based approaches have been developed to improve the performance of multi-object tracking by applying the most accurate and efficient combinations of object detection models and appearance embedding extraction models. However, two-stage methods show a low inference speed since the embedding extraction can only be performed at the end of the object detection. To alleviate this problem, single-shot methods, which simultaneously perform object detection and embedding extraction, have been developed and have drastically improved the inference speed. However, there is a trade-off between accuracy and efficiency. Therefore, this study proposes an enhanced single-shot multi-object tracking system that displays improved accuracy while maintaining a high inference speed. With a strong feature extraction and fusion, the object detection of our model achieves an AP score of 69.93% on the UA-DETRAC dataset and outperforms previous state-of-the-art methods, such as FairMOT and JDE. Based on the improved object detection performance, our multi-object tracking system achieves a MOTA score of 68.5% and a PR-MOTA score of 24.5% on the same dataset, also surpassing the previous state-of-the-art trackers.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Efficient Single-Shot Multi-Object Tracking for Vehicles in Traffic Scenarios

Lee

Yoo

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…Mov-ing away from having a central server also avoids having a single point of failure. For example, crash detection in traffic surveillance [148], invalid router updates in network connections [103], pedestrian detection in self-driving cars [215], anomalous condition detection in medical electrocardiograms [30], etc. are all mission critical scenarios which require quick decision making and tractability.…”

Section: Introductionmentioning

confidence: 99%

“…Detecting accidents quickly can enable emergency services to be deployed in time. There are a few studies which have tackled this problem [13,45,148]. The method by Doshi et al [45] is fast but it does not capture the motion of the object.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Framework for Anomaly Detection under Dynamic and Distributed Scenarios

Odiathevar¹

View full text Add to dashboard Cite

Anomaly Detection is an important aspect of many application domains. It refers to the problem of finding patterns in data that do not conform to expected behaviour. Hence, understanding of expected behaviour well is fundamental to performing effective anomaly detection. However, data profiles constantly evolve in certain domains such as computer networks. In other domains such as traffic monitoring and healthcare, data are distributed and are either too large or there are privacy concerns in transmitting them to a central location. These situations pose a challenge to obtain an accurate understanding of non-anomalous profiles. Changing profiles undermine existing anomaly detection models and make them less effective. Training a robust model with data from multiple sources is also challenging. Moreover, in real world scenarios, it is not apparent how an anomaly detection model can be built to address the problem. This thesis focuses on the building of a robust anomaly detection system where data profiles evolve and/or are distributed. It proposes a novel Online Offline Framework to separate existing expected behaviour, new possible expected behaviour and anomalies in streaming data. It also addresses the distributed scenario using a theoretically sound fully Bayesian approach. These methods improve performances of anomaly detection systems and work well with biased and uneven data partitions. The proposed methods are validated using real world data in three different domains. This thesis identifies the implementation difficulties in these domains and produces three novel methodologies to address each of the core anomaly detection problems.

show abstract