Fast single shot multibox detector and its application on vehicle counting system

Chen, Lili; Zhang, Zhengdao

doi:10.1049/iet-its.2018.5005

Cited by 27 publications

(16 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[29][30][31] In existing researches, some scholars also apply CNNs to detect vehicles. Chen et al 35 proposed a real-time multi-type vehicle counting system based on SSD. Zhang et al 36 suggested a vehicle detection method that is usable in complex scenes by improving Faster-RCNN architecture.…”

Section: Yolo-v3-based Dual-target Detection Model For the Entire Bmentioning

confidence: 99%

An accurate and robust monitoring method of full‐bridge traffic load distribution based on YOLO‐v3 machine vision

Dan

2020

Struct Control Health Monit

View full text Add to dashboard Cite

The accurate and stable identification of the traffic load distribution on the bridge deck is of great significance to bridge health monitoring and safety early warning. To accomplish this task, we have combined the weigh-in-motion system (WIMs) with machine vision and developed a traffic load monitoring (TLM) technology for the whole bridge deck. For bridge health monitoring, the TLM should be available for online structural analysis, have high accuracy, and be able to adapt to changes in lighting conditions. However, existing TLM methods are difficult to meet the requirements of real-time, accuracy, and lighting robustness simultaneously. In this regard, this paper proposes an improved full-bridge TLM method based on YOLO-v3 convolutional neural network. The core of this method includes training a dual-target detection model and correcting vehicle locations. The detection model can identify profiles of the entire vehicle and its tail and can mark them with compact rectangular boxes. Based on the corner points of these rectangular boxes, an optical geometry model is proposed to measure vehicle dimensions and correct vehicle centroids, thereby the vehicle locations can be estimated more accurately. By applying the time synchronization of cameras and the WIMs, each measured load is paired with the vehicle "pixel cluster" detected in the video; further, the traffic load distribution on the whole bridge deck is identified accurately in real-time. Verified by the field data of a ramp bridge, the proposed method is proved more accurately on the identification of vehicle locations, more robust lighting adaptability, and faster calculation speed, which can meet the requirements of field monitoring of traffic load distribution.

show abstract

Section: Yolo-v3-based Dual-target Detection Model For the Entire Bmentioning

confidence: 99%

An accurate and robust monitoring method of full‐bridge traffic load distribution based on YOLO‐v3 machine vision

Dan

2020

Struct Control Health Monit

View full text Add to dashboard Cite

show abstract

“…SSD has been used for vehicle detection in several studies; e.g. in its original form [38], with a modified architecture combined with the Slim ResNet-34 [39], aided with temporally identified regions of interest [40]. YOLO and RetinaNet architectures are also used in the literature for vehicle detection.…”

Section: Related Workmentioning

confidence: 99%

“…These approaches are performed on relatively low-resolution videos in order to achieve high detection speed and proper accuracy. Examples include resolution of 960 × 540 pixels at 34 fps with average precision of 43%-79% [38], resolution of 300 × 300 pixels at 20 fps with mean average precision (mAP) of 77% [39], varying resolutions up to 1920 × 1080 pixels with processing time of 0.09 sec per frame (~11 fps) with F1-score of 39% [40], varying resolutions up to 608 × 608 pixels with processing time of 0.038 sec per frame (~26 fps) with mAP of 68% [42], resolution of 960 × 540 pixels at 9 fps with mAP of 85% [43], resolution of 512 × 512 pixels at 75 fps with mAP of 80%-89% [44], resolution of 960 × 540 pixels at 21 fps with mAP of 74% [45]. A comprehensive report on the performance of single-network detectors against two-stage networks can be found in [46].…”

Section: Related Workmentioning

confidence: 99%

Orientation- and Scale-Invariant Multi-Vehicle Detection and Tracking from Unmanned Aerial Videos

2019

View full text Add to dashboard Cite

Along with the advancement of light-weight sensing and processing technologies, unmanned aerial vehicles (UAVs) have recently become popular platforms for intelligent traffic monitoring and control. UAV-mounted cameras can capture traffic-flow videos from various perspectives providing a comprehensive insight into road conditions. To analyze the traffic flow from remotely captured videos, a reliable and accurate vehicle detection-and-tracking approach is required. In this paper, we propose a deep-learning framework for vehicle detection and tracking from UAV videos for monitoring traffic flow in complex road structures. This approach is designed to be invariant to significant orientation and scale variations in the videos. The detection procedure is performed by fine-tuning a state-of-the-art object detector, You Only Look Once (YOLOv3), using several custom-labeled traffic datasets. Vehicle tracking is conducted following a tracking-by-detection paradigm, where deep appearance features are used for vehicle re-identification, and Kalman filtering is used for motion estimation. The proposed methodology is tested on a variety of real videos collected by UAVs under various conditions, e.g., in late afternoons with long vehicle shadows, in dawn with vehicles lights being on, over roundabouts and interchange roads where vehicle directions change considerably, and from various viewpoints where vehicles’ appearance undergo substantial perspective distortions. The proposed tracking-by-detection approach performs efficiently at 11 frames per second on color videos of 2720p resolution. Experiments demonstrated that high detection accuracy could be achieved with an average F1-score of 92.1%. Besides, the tracking technique performs accurately, with an average multiple-object tracking accuracy (MOTA) of 81.3%. The proposed approach also addressed the shortcomings of the state-of-the-art in multi-object tracking regarding frequent identity switching, resulting in a total of only one identity switch over every 305 tracked vehicles.

show abstract

“…To realize the above system, a technology for detecting objects from the running image and a technology to construct the contributing model for selecting the contributing objects are required. For the former, object detection technologies based on deep learning, such as Single Shot multi-box Detector (SSD) [14][15][16] or YOLO, [17][18][19] have been proposed. For the feature extraction, approaches such as Variational AutoEncoder [20][21][22] have been proposed.…”

Section: Preliminariesmentioning

confidence: 99%

Detection of contributing object to driving operations based on hidden Markov model

Hashimoto¹,

Yamada²,

Tsuchiya³

et al. 2019

International Journal of Advanced Robotic Systems

View full text Add to dashboard Cite

With increase in the number of elderly people in the Japanese society, traffic accidents caused by elderly driver is considered problematic. The primary factor of the traffic accidents is a reduction in their driving cognitive performance. Therefore, a system that supports the cognitive performance of drivers can greatly contribute in preventing accidents. Recently, the development of devices for visually providing information, such as smart glasses or head up display, is in progress. These devices can provide more effective supporting information for cognitive performance. In this article, we focus on the selection problem of information to be presented for drivers to realize the cognitive support system. It has been reported that the presentation of excessive information to a driver reduces the judgment ability of the driver and makes the information less trustworthy. Thus, indiscriminate presentation of information in the vision of the driver is not an effective cognitive support. Therefore, a mechanism for determining the information to be presented to the driver based on the current driving situation is required. In this study, the object that contributes to execution of avoidance driving operation is regarded as the object that drivers must recognize and present for drivers. This object is called as contributing object. In this article, we propose a method that selects contributing objects among the appeared objects on the current driving scene. The proposed method expresses the relation between the time series change of an appeared object and avoidance operation of the driver by a mathematical model. This model can predict execution timing of avoidance driving operation and estimate contributing object based on the prediction result of driving operation. This model named as contributing model consisted of multi-hidden Markov models. Hidden Markov model is time series probabilistic model with high readability. This is because that model parameters express the probabilistic distribution and its statistics. Therefore, the characteristics of contributing model are that it enables the designer to understand the basis for the output decision. In this article, we evaluated detection accuracy of contributing object based on the proposed method, and readability of contributing model through several experiments. According to the results of these experiments, high detection accuracy of contributing object was confirmed. Moreover, it was confirmed that the basis of detected contributing object judgment can be understood from contributing model.

show abstract

Fast single shot multibox detector and its application on vehicle counting system

Cited by 27 publications

References 24 publications

An accurate and robust monitoring method of full‐bridge traffic load distribution based on YOLO‐v3 machine vision

An accurate and robust monitoring method of full‐bridge traffic load distribution based on YOLO‐v3 machine vision

Orientation- and Scale-Invariant Multi-Vehicle Detection and Tracking from Unmanned Aerial Videos

Detection of contributing object to driving operations based on hidden Markov model

Contact Info

Product

Resources

About