Search results

The purpose of the study is to address the problems of low accuracy and missed detection of occluded pedestrians and small target pedestrians when using the YOLOX general object detection algorithm for pedestrian detection. This study proposes a multi-level fine-grained YOLOX pedestrian detection algorithm.

First, to address the problem of the original YOLOX algorithm in obtaining a single perceptual field for the feature map before feature fusion, this study improves the PAFPN structure by adding the ResCoT module to increase the diversity of the perceptual field of the feature map and divides the pedestrian multi-scale features into finer granularity. Second, for the CSPLayer of the PAFPN, a weight gain-based normalization-based attention module (NAM) is proposed to make the model pay more attention to the context information when extracting pedestrian features and highlight the salient features of pedestrians. Finally, the authors experimentally determined the optimal values for the confidence loss function.

The experimental results show that, compared with the original YOLOX algorithm, the AP of the improved algorithm increased by 2.90%, the Recall increased by 3.57%, and F1 increased by 2% on the pedestrian dataset.

The multi-level fine-grained YOLOX pedestrian detection algorithm can effectively improve the detection of occluded pedestrians and small target pedestrians.

The authors introduce a multi-level fine-grained ResCoT module and a weight gain-based NAM attention module.

The purpose of this paper is to summarize the existing point cloud target detection algorithms based on deep learning, and provide reference for researchers in related fields. In recent years, with its outstanding performance in target detection of 2D images, deep learning technology has been applied in light detection and ranging (LiDAR) point cloud data to improve the automation and intelligence level of target detection. However, there are still some difficulties and room for improvement in target detection from the 3D point cloud. In this paper, the vehicle LiDAR target detection method is chosen as the research subject.

Firstly, the challenges of applying deep learning to point cloud target detection are described; secondly, solutions in relevant research are combed in response to the above challenges. The currently popular target detection methods are classified, among which some are compared with illustrate advantages and disadvantages. Moreover, approaches to improve the accuracy of network target detection are introduced.

Finally, this paper also summarizes the shortcomings of existing methods and signals the prospective development trend.

This paper introduces some existing point cloud target detection methods based on deep learning, which can be applied to a driverless, digital map, traffic monitoring and other fields, and provides a reference for researchers in related fields.

Biometric identification system has become emerging research field because of its wide applications in the fields of security. This study (multimodal system) aims to find more applications than the unimodal system because of their high user acceptance value, better recognition accuracy and low-cost sensors. The biometric identification using the finger knuckle and the palmprint finds more application than other features because of its unique features.

The proposed model performs the user authentication through the extracted features from both the palmprint and the finger knuckle images. The two major processes in the proposed system are feature extraction and classification. The proposed model extracts the features from the palmprint and the finger knuckle with the proposed HE-Co-HOG model after the pre-processing. The proposed HE-Co-HOG model finds the Palmprint HE-Co-HOG vector and the finger knuckle HE-Co-HOG vector. These features from both the palmprint and the finger knuckle are combined with the optimal weight score from the fractional firefly (FFF) algorithm. The layered k-SVM classifier classifies each person's identity from the fused vector.

Two standard data sets with the palmprint and the finger knuckle images were used for the simulation. The simulation results were analyzed in two ways. In the first method, the bin sizes of the HE-Co-HOG vector were varied for the various training of the data set. In the second method, the performance of the proposed model was compared with the existing models for the different training size of the data set. From the simulation results, the proposed model has achieved a maximum accuracy of 0.95 and the lowest false acceptance rate and false rejection rate with a value of 0.1.

In this paper, the multimodal biometric recognition system based on the proposed HE-Co-HOG with the k-SVM and the FFF is developed. The proposed model uses the palmprint and the finger knuckle images as the biometrics. The development of the proposed HE-Co-HOG vector is done by modifying the Co-HOG with the holoentropy weights.

The purpose of this study is to propose a method for vulnerable pedestrians to visualize potential obstacles on sidewalks. In recent years, the number of vulnerable pedestrians has been increasing as Japanese society has aged. The number of wheelchair users is also expected to increase in the future. Currently, barrier-free maps and street-view applications can be used by wheelchair users to check possible routes and the surroundings of their destinations in advance. However, identifying physical barriers that pose a threat to vulnerable pedestrians en route is often difficult.

This study uses photogrammetry to create a digital twin of the three-dimensional (3D) geometry of the existing walking space by collecting photographic images taken on sidewalks. This approach allows for the creation of high-resolution digital elevation models of the entire physical sidewalk surface from which physical barriers such as local gradients and height differences can be detected by uniform image filtering. The method can be used with a Web-based data visualization tool in a geographical information system, permitting first-person views of the ground and accurate geolocation of the barriers on the map.

The findings of this study showed that capturing the road surface with a small wide-angle camera while walking is sufficient for recording subtle 3D undulations in the road surface. The method used for capturing data and the precision of the 3D restoration results are described.

The proposed approach demonstrates the significant benefits of creating a digital twin of walking space using photogrammetry as a cost-effective means of balancing the acquisition of 3D data that is sufficiently accurate to show the detailed geometric features needed to navigate a walking space safely. Further, the findings showed how information can be provided directly to users through two-dimensional (2D) and 3D Web-based visualizations.

The purpose of this paper is to describe recent progress on the “Crosswatch” project, a smartphone‐based system developed for providing guidance to blind and visually impaired travelers at traffic intersections. Building on past work on Crosswatch functionality to help the user achieve proper alignment with the crosswalk and read the status of walk lights to know when it is time to cross, the authors outline the directions Crosswatch is now taking to help realize its potential for becoming a practical system: namely, augmenting computer vision with other information sources, including geographic information systems (GIS) and sensor data, and inferring the user's location much more precisely than is possible through GPS alone, to provide a much larger range of information about traffic intersections to the pedestrian.

The paper summarizes past progress on Crosswatch and describes details about the development of new Crosswatch functionalities. One such functionality, which is required for determination of the user's precise location, is studied in detail, including the design of a suitable user interface to support this functionality and preliminary tests of this interface with visually impaired volunteer subjects.

The results of the tests of the new Crosswatch functionality demonstrate that the functionality is feasible in that it is usable by visually impaired persons.

While the tests that were conducted of the new Crosswatch functionality are preliminary, the results of the tests have suggested several possible improvements, to be explored in the future.

The results described in this paper suggest that the necessary technologies used by the Crosswatch system are rapidly maturing, implying that the system has an excellent chance of becoming practical in the near future.

The paper addresses an innovative solution to a key problem faced by blind and visually impaired travelers, which has the potential to greatly improve independent travel for these individuals.

The purpose of this paper is to propose a robust iterative LIDAR-based pose tracking method assisted by modified visual odometer to resist initial value disturbance and locate a robot in the environments with certain occlusion.

At first, an iterative LIDAR-based pose tracking method is proposed. The LIDAR information is filtered and occupancy grid map is pre-processed. The sample generation and scoring are iterated so that the result is converged to the stable value. To improve the efficiency of sample processing, the integer-valued map indices of rotational samples are preserved and translated. All generated samples are analyzed to determine the maximum error direction. Then, a modified visual odometer is introduced for error compensation. The oriented fast and rotated brief (ORB) features are uniformly sampled in the image. A local map which contains key frames for reference is maintained. These two measures ensure that the modified visual odometer is able to return robust result which compensates the error of LIDAR-based pose tracking method in the maximum error direction.

Three experiments are conducted to prove the advantages of the proposed method. The proposed method can resist initial value disturbance with high computational efficiency, give back credible real-time result in the environment with abundant features and locate a robot in the environment with certain occlusion.

The proposed method is able to give back real-time pose tracking results with robustness. The iterative sample generation enables the robot to resist initial value disturbance. In each iteration, rotational and translational samples are separately generated to enhance computational efficiency. The maximum error direction of LIDAR-based pose tracking method is determined by principle component analysis and compensated by the result of modified visual odometer to give back correct pose in the environment with certain occlusion.

The purpose of this paper is to propose a defect detection method of silicone caps positional deviation on flexible printed circuit board (FPCB) of keyboard based on automatic optical inspection.

First, the center of silicone caps of target keyboard FPCB image was extracted as feature points for generating the feature image which is used for registration rigidly with the reference feature image generated from the CAD drawings. Then, a flexible image registration method based on the surrounding-control-center B-splines (SCCB) strategy was proposed, which could correct the flexible deformation of the image generated by FPCB substrate while keeping the pasting deviation information about silicone caps unchanged. Finally, on this basis, a nearest neighbor strategy was proposed to detect the positional deviation of silicone caps.

Experimental results show that the proposed method can effectively detect the positional deviation defect of silicone caps. The G-mean value of the proposed method is 0.941746, which is 0.3 higher compared to that of similar research.

This paper presents a method to detect positional deviation defect of silicone caps on keyboard FPCB. Different from the classic B-spline image registration method, the proposed SCCB method used the neighborhood information of the pixel to be registered selectively to calculate the displacement vector needed for its registration, which overcame the problem that the silicone cap pasting deviation information disappears with the correction of the flexible deformation of the image.

One of the most neglected sources of energy loss is streetlights that generate too much light in areas where it is not required. Energy waste has enormous economic and environmental effects. In addition, due to the conventional manual nature of operation, streetlights are frequently seen being turned ‘ON’ during the day and ‘OFF’ in the evening, which is regrettable even in the twenty-first century. These issues require automated streetlight control in order to be resolved. This study aims to develop a novel streetlight controlling method by combining a smart transport monitoring system powered by computer vision technology with a closed circuit television (CCTV) camera that allows the light-emitting diode (LED) streetlight to automatically light up with the appropriate brightness by detecting the presence of pedestrians or vehicles and dimming the streetlight in their absence using semantic image segmentation from the CCTV video streaming. Consequently, our model distinguishes daylight and nighttime, which made it feasible to automate the process of turning the streetlight ‘ON’ and ‘OFF’ to save energy consumption costs. According to the aforementioned approach, geo-location sensor data could be utilised to make more informed streetlight management decisions. To complete the tasks, we consider training the U-net model with ResNet-34 as its backbone. Validity of the models is guaranteed with the use of assessment matrices. The suggested concept is straightforward, economical, energy-efficient, long-lasting and more resilient than conventional alternatives.

It is an important style classification way to sort out suits according to the button arrangement. However, since the different dressing ways of suit cause the buttons to be easily occluded, the traditional identification methods are difficult to identify the details of suits, and the recognition accuracy is not ideal. The purpose of this paper is to solve the problem of fine-grained classification of suit by button arrangement. Taking men's suits as an example, a method of coordinate position discrimination algorithm combined faster region-based convolutional neural network (R-CNN) algorithm is proposed to achieve accurate batch classification of suit styles under different dressing modes.

The detection algorithm of suit buttons proposed in this paper includes faster R-CNN algorithm and coordinate position discrimination algorithm. Firstly, a small sample base was established, which includes six suit styles in different dressing states. Secondly, buttons and buttonholes in the image were marked, and the image features were extracted by the residual network to identify the object. The anchors regression coordinates in the sample were obtained through convolution, pooling and other operations. Finally, the position coordinate relation of buttons and buttonholes was used to accurately judge and distinguish suit styles under different dressing ways, so as to eliminate the wrong results of direct classification by the network and achieve accurate classification.

The experimental results show that this method could be used to accurately classify suits based on small samples. The recognition accuracy rate reaches 95.42%. It can effectively solve the problem of machine misjudgment of suit style due to the cover of buttons, which provides an effective method for the fine-grained classification of suit style.

A method combining coordinate position discrimination algorithm with convolutional neural network was proposed for the first time to realize the fine-grained classification of suit style. It solves the problem of machine misreading, which is easily caused by buttons occluded in different suits.