Figure - available from: Neural Computing and Applications
This content is subject to copyright. Terms and conditions apply.
Source publication
Convolutional neural network has shown strong capability to improve performance in vehicle detection, which is one of the main research topics of intelligent transportation system. Aiming to detect the blocked vehicles efficiently in actual traffic scenes, we propose a novel convolutional neural network based on multi-target corner pooling layers....
Similar publications
Photo- and video-realistic generation techniques have become a reality following the advent of deep neural networks. Consequently, there are immense concerns regarding the difficulty in differentiating what content is real from what is synthetic. An example of video-realistic generation techniques is the infamous Deepfakes, which exploit the main m...
Citations
... Vehicle object detection is an important application in the field of computer vision, which has developed rapidly in recent years, exhibiting broad prospects [8][9][10]. Currently, the research on target detection algorithms is mainly divided into two categories: object detection algorithms based on traditional image processing and machine learning algorithms, and object detection algorithms based on deep learning [11,12]. ...
Vehicle detection is a very important part in intelligent transportation system. In order to improve the detection speed without sacrificing the accuracy, this paper propose an improved Faster R-CNN algorithm based on frame difference and spatiotemporal context to realize real-time detection of vehicles. We improve the training and testing speed of Faster R-CNN by improving the RPN module. Different from the original Faster R-CNN’s anchor based strategy, the inter frame difference in this paper is mainly used to extract the region of interest of the target. And we introduce spatiotemporal context information to assist our detection. Among them, the spatial context is formed by adding the association information outside the target area to enhance the expression of target information and improve the accuracy of target detection, while the anchor filtering of the original Faster R-CNN can be carried out by integrating the temporal context information, so as to improve the detection efficiency. This improved RPN has strong pertinence to the detection of moving vehicles. This strategy not only makes our branch network parallelly process with the original Faster R-CNN, but also avoids the extra time consumption caused by the addition of algorithm. More importantly, it can be simply added to the existing Faster R-CNN based application system without algorithm adjustment or network retraining. Experimental results show that the proposed method has high detection efficiency and low sensitivity to background changes.
... Semantic segmentation has recently witnessed great progress driven by the advancement of Convolutional Neural Networks (CNNs) [5], especially Fully Convolutional Networks (FCNs) [6]. Thanks to deeply learned features [7] and large-scale annotations [8], U-Net [9], a CNN-based Network, has become the state-of-the-art technology for human image segmentation. ...
Human image segmentation has been a practical and active research topic due to its wide range of potential application. There are some previous studies on manual, semi-automatic and automatic segmentation methods to investigate the semantic segmentation of human parts fully for real-world human analysis scenarios, but further research is still needed. This paper presents a novel semantic segmentation network, named TRCA-Net, for human image segmentation tasks. Having the TransUNet as the backbone, TRCA-Net incorporates Res2Net and Coordinate Attention to improve the performance. Res2Net blocks and Transformer can obtain better feature maps by encoding the input images. The Coordinate Attention in the decoder aggregates and upsamples the encoded feature maps, and connects to the high-resolution CNN feature maps for gaining accurate segmentation. The TRCA-Net can enhance finer details by recovering local spatial information. We compare the TRCA-Net with state-of-the-art (SOAT) semantic segmentation networks: the original U-Net, DeepLabv3+, and TransUNet. The experiment results have demonstrated that our proposed TRCA-Net outperforms these networks.
... ITS processes the information acquired by surveillance cameras to automate traffic monitoring, thereby guiding the traffic authorities to take advert possible mishaps caused by commuters. The acquired surveillance footage is useful in ITS to carry out vehicle detection [4] [5], vehicle counting [6], vehicle re-identification [1] and tracking [7] [8]. Currently, more efforts are put in place in computer vision to develop a robust vehicle reidentification framework. ...
Vehicle re-identification is an important feature of an intelligent transportation system as part of a smart city application. Vehicle re-identification aims at matching vehicles from images acquired by surveillance cameras at different locations. During rush hours, vehicles are densely occupied across regions such as entry/exit of gated campuses, railways, airports, educational institutions, etc. Due to this uneven flow of traffic, there is a possibility of violation of traffic rules by the vehicles that lead to a security breach. In such scenarios to speed up the re-identification process, it is justified to look into a specific group of surveillance cameras to detect and re-identify vehicles on day to day basis in near real-time. However, the existing vehicle re-identification datasets do not contain zone specific information and therefore can not be used to evaluate the performance of re-identification algorithms in different zones. In the proposed work for re-identification, a framework is developed that performs vehicle re-identification across a group of cameras that monitors varying traffic movements over an area. These areas defined as “strategic zones” comprise a subset of non-overlapping cameras that are installed to monitor non-uniform vehicle movements. The re-identification framework is evaluated on a novel dataset developed to understand the performance of vehicle re-identification across strategic zones. The dataset consists of videos of vehicles captured through 20 CCTV surveillance cameras that are grouped into four different zones. Various experiments are conducted to study the performance of re-identification across four zones using a deep neural network with triplet loss, L2 regularization, and re-ranking. The experiments conducted with an image dimension of 224 × 224 have demonstrated an overall mAP of 77.22%. Also, for each of the four zones a mAP of 82.16%, 69.1%, 66.5%, and 75.76% is achieved. The experimental results demonstrate huge variations in the accuracy of vehicle re-identification method across different zones. Therefore, the study assess the possible measures that can be taken to improve the performance in individual zones for an accurate vehicle re-identification in intelligent transport system.
... These extraction methods have obvious limitations, and the credibility of the calculation results is lower. Deep learning-based methods [12][13][14] show their advantages in image classification [15], target detection [16], etc. The models proposed in [13,17,18] have successfully verified the ability of the approach primarily based on deep convolutional neural network (DCNN) for studying features. ...
... The EIIs extract image features by using a coding method based on DCNN to learn binary hash functions. Preprocessing diagram is shown in Fig. 4. Firstly, images are identified by recognition technology [16] to determine the main object, and feature vectors of images are obtained by convolution layer; chunking layer performs block processing on the vector to obtain multiple sub-vectors; each sub-vector passes through fully connected layer and activation layer in turn, then enters merging layer; merging layer merges multiple sub-vectors into one vector; after the vector enters thresholding layer, the image feature D ¼ fx 1 ; x 2 ; Á Á Á x n g is obtained. Preprocess of the detailed steps is shown in Sect. ...
With 5G and Internet technologies developing rapidly, outsourcing images to cloud servers has attracted growing attention. In existing technologies, images are often outsourced to cloud servers to reduce storage and computing burdens. However, outsourcing images to cloud servers without any processing may reveal the users’ privacy, because the images may contain sensitive information about users, such as faces and locations, especially in electronic investigation. To overcome the security problems in image retrieval, we propose a privacy-preserving image retrieval scheme based on deep convolutional neural network (DCNN) and vector homomorphic encryption (VHE). We adopt DCNN and hash algorithms to extract image feature vectors, which improves retrieval accuracy. By combining VHE and K-means outsourcing clustering algorithms, the cloud server can build encrypted index trees, which speeds up the search and reduces the computational cost. In addition, a lightweight access control technique is used to allow image owners to set access policies for datasets flexibly. We prove the security of the proposed scheme and show the effectiveness of the scheme through experiments. Our scheme is suitable for application in electronic image investigation systems (EIIs) to optimize the storage and search of police data.
... The corner points of the vehicle are one of the main important features in the detection of the vehicle. The authors Li-Vang Hao et al. [10], have proposed an approach based on a convolution neural network that depends on multi-target corner pooling layers. The missed corner points are generated by the network helps in improving the accuracy. ...
The paper proposes a robust approach to detect and track the vehicle under various climatic conditions and in the presence of camera shake, shadows, sudden illumination change. Corners have significant features to detect and track the vehicle. Corner points from the vehicular region are segmented from non - vehicular regions based on the statistical background corner point model. The foreground corner points that belong to the vehicular region are grouped using Euclidean distance as they are closely associated with each other. The flickering effects caused by the corner detection algorithm are handled by tracking these corner points. The detection accuracy of the algorithm is 94.32%.
... In [10], a new neural network architecture based on multi-target corner pooling layer (MTCP) is proposed to detect a car. This method allows effectively to solve the problem of detecting the vehicles which are partially overlapped on the image, but it is affected by weather and light. ...
A new mathematical model of the intelligent access monitoring and control system based on the cybernetic approach is proposed for solving the problems of vehicle access to the territory of an organization. The distinctive feature of the mathematical model is the ability to take into account and recognize normal and abnormal situations at the protected object and develop control actions. To localize vehicles and recognize their license plates, the composition of traditional methods of image processing and two-pass classification performed by the developed modified architecture of convolutional neural network MobileNet is offered. The composition allows to define additional identification features of the object. The proposed adaptations allow to recognize the situation in real time with low computational costs and high accuracy. The natural experiment has shown that the integration of the modern hardware means and algorithms of object detection and recognition, even in the rough conditions of street closed-circuit television monitoring, provides not less than 96% accuracy, and the processing time of one frame is not more than 0.094 s based on the Nvidia GeForce 1080Ti graphic processor. High recognition accuracy without loss of speed in the real-time mode is achieved by integrating the modern hardware means and the algorithms of object detection and recognition. The program module in Python using the Tensorflow and Keras library is developed for carrying out the access control functions.
Multi-target detection based on corner pooling provides a distinctive framework without anchor boxes, which has achieved wide application in the area of intelligent transportation system. To effectively detect small vehicles in the distant view, we propose an improved detection network termed corner pooling with attention mechanism (CPAM). A newly devised network called Hourglass with Coordinate Attention(Hourglass-CA) is proposed as an alternative to the Hourglass-104 backbone network. This one incorporates a multi-level attention mechanism to optimize the efficiency of feature extraction. Additionally, a novel multi-level attention loss(MLA loss) is presented, which dynamically adjusts the offsets during the feature extraction process. The experimental results demonstrate that our proposed CPAM achieves lightweight detection, reducing the parameters from 201M to 117M with an FPS from 4.2 to 16.1. Moreover, the AP can reach 51.6%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%$$\end{document}, surpassing several existing detectors.
Graphical abstract
Stack Overflow provides a platform for developers to seek suitable solutions by asking questions and receiving answers on various topics. However, many questions are usually not answered quickly enough. Since the questioners are eager to know the specific time interval at which a question can be answered, it becomes an important task for Stack Overflow to feedback the answer time to the question. To address this issue, we propose a model for predicting the answer time of questions, named Predicting Answer Time (i.e., PAT model), which consists of two parts: a feature acquisition and fusion model, and a deep neural network model. The framework uses a variety of features mined from questions in Stack Overflow, including the question description, question title, question tags, the creation time of the question, and other temporal features. These features are fused and fed into the deep neural network to predict the answer time of the question. As a case study, post data from Stack Overflow are used to assess the model. We use traditional regression algorithms as the baselines, such as Linear Regression, K-Nearest Neighbors Regression, Support Vector Regression, Multilayer Perceptron Regression, and Random Forest Regression. Experimental results show that the PAT model can predict the answer time of questions more accurately than traditional regression algorithms, and shorten the error of the predicted answer time by nearly 10 hours.
