Finding every car: a traffic surveillance multi-scale vehicle object detection method

Mao, Qi-Chao; Sun, Hong-Mei; Zuo, Ling-Qun; Jia, Rui-Sheng

doi:10.1007/s10489-020-01704-5

Cited by 34 publications

(12 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, the YOLOv2+DarkNet19 built by Sang et al [51] was more accurate than the others, but the speed is significantly lower than the SSD+VGG16 built by Dai et al [50]. Moreover, the accuracy of the results may not necessarily improve with the increasing of training sets, and suitability may be more important, such as SSD+ResNet [49] and SSD+VGG16 [50], YOLOv2+DarkNet19 [51] and YOLOv3+DarkNet53 [52].…”

Section: B Comparison With Recent State-of-the-art Methodsmentioning

confidence: 98%

A Video-Based Method With Strong-Robustness for Vehicle Detection and Classification Based on Static Appearance Features and Motion Features

Chen

2021

IEEE Access

View full text Add to dashboard Cite

Vehicle detection and classification plays an important role in intelligent transportation system. Compared with traditional detectors, the detection and classification based on traffic surveillance video shows a huge advantage in its flexibility and continuity. However, to get wide applicability and strong robustness, most current methods focus on improving the accuracy of detectors by adjusting network parameters constantly, or increasing the size of training sets, which challenges the collection and labeling of data, the performance of computers, the scope of application and so on. Moreover, the unique continuity characteristic of the video, which can be used to describe the motion features of vehicle, is often ignored. Take these facts into account, this paper proposed a video-based vehicle detection and classification method, which is based on static appearance features and motion features both. Four detectors of different performance were trained with small training sets, and the designed algorithms for the remove, selection and reorganization of detected objects contribute to obtaining the optimal results of detection and classification. The experiment results show that the proposed method is able to detect and classify vehicles with more than 0.95 accuracy dealing with different road environments.

show abstract

Section: B Comparison With Recent State-of-the-art Methodsmentioning

confidence: 98%

A Video-Based Method With Strong-Robustness for Vehicle Detection and Classification Based on Static Appearance Features and Motion Features

Chen

2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Finally, the number of channels is restored to 1C through 1 1  convolution, and the features are accumulated with the output features of the secondly step. In order to reduce the loss of accuracy caused by float16 inference on embedded platforms, RELU6 is used as the activation function of the first three convolutional layers [42]. At the same time, the linear activation function is used as the output of the last layer.…”

Section: Down-sampling Inverted Residual Blockmentioning

confidence: 99%

A Surveillance Video Real-Time Analysis System Based on Edge-Cloud and FL-YOLO Cooperation in Coal Mine

Zhang

2021

IEEE Access

View full text Add to dashboard Cite

Video monitoring is an important means to ensure production safety in coal mine. However, the currently intelligent video surveillance is difficult to respond in real-time due to the latency of cloud computing. In this paper, a cloud-edge cooperation framework is proposed, which integrates cloud computing and edge computing in a coordinated manner. The cloud computing is used to process non-realtime and global tasks, while the edge computing is responsible for handling local monitoring video in realtime. In order to realize cloud-edge data interaction and online optimization for edge models, the heterogeneous converged network is built. In addition, an object detection model FL-YOLO composed of depthwise separable convolution and down-sampling inverted residual block is proposed, which realizes real-time video analysis at the edge. Finally, this paper discusses the complexity of FL-YOLO by its computational cost and model size. The experiment results show that the model size of FL-YOLO is 16.1MB, which is very light, and it achieves 36.7 FPS on NVIDIA Jetson TX1 and an AP of 76.7% on Multi-scene pedestrian dataset. Comparing with mainstream object detection models, FL-YOLO completes faster detection speed and higher accuracy, and it has lower calculation complexity and smaller model scale. Furthermore, the AP on Single-scene pedestrian dataset of FL-YOLO is improved to 80.7% by cloud-edge cooperation. K-Fold method is also used to further compared the performance of FL-YOLO and other models. Moreover, system test is implemented on coal mine, which validates the actual engineering effect of the proposed cloud-edge cooperation framework.INDEX TERMS Edge computing, YOLO, cloud-edge cooperation, real-time analysis.

show abstract

“…The controller algorithm is then applied to mitigate collisions in a real-time scenario. Mao et al [35] addressed the multi-scale vehicle detection, and the overlapping objects issues in autonomous vehicle settings. It extended the YOLOv3 by improving the feature extraction process while using the inverted residuals strategy on the convolution layers.…”

Section: Image Searchmentioning

confidence: 99%

Deep learning based decomposition for visual navigation in industrial platforms

et al. 2021

View full text Add to dashboard Cite

In the heavy asset industry, such as oil & gas, offshore personnel need to locate various equipment on the installation on a daily basis for inspection and maintenance purposes. However, locating equipment in such GPS denied environments is very time consuming due to the complexity of the environment and the large amount of equipment. To address this challenge we investigate an alternative approach to study the navigation problem based on visual imagery data instead of current ad-hoc methods where engineering drawings or large CAD models are used to find equipment. In particular, this paper investigates the combination of deep learning and decomposition for the image retrieval problem which is central for visual navigation. A convolutional neural network is first used to extract relevant features from the image database. The database is then decomposed into clusters of visually similar images, where several algorithms have been explored in order to make the clusters as independent as possible. The Bag-of-Words (BoW) approach is then applied on each cluster to build a vocabulary forest. During the searching process the vocabulary forest is exploited to find the most relevant images to the query image. To validate the usefulness of the proposed framework, intensive experiments have been carried out using both standard datasets and images from industrial environments. We show that the suggested approach outperforms the BoW-based image retrieval solutions, both in terms of computing time and accuracy. We also show the applicability of this approach on real industrial scenarios by applying the model on imagery data from offshore oil platforms.

show abstract

Finding every car: a traffic surveillance multi-scale vehicle object detection method

Cited by 34 publications

References 33 publications

A Video-Based Method With Strong-Robustness for Vehicle Detection and Classification Based on Static Appearance Features and Motion Features

A Video-Based Method With Strong-Robustness for Vehicle Detection and Classification Based on Static Appearance Features and Motion Features

A Surveillance Video Real-Time Analysis System Based on Edge-Cloud and FL-YOLO Cooperation in Coal Mine

Deep learning based decomposition for visual navigation in industrial platforms

Contact Info

Product

Resources

About