Single Shot MultiBox Detector Algorithm Research Articles

Traffic Sign Detection Based on SSD Combined with Receptive Field Module and Path Aggregation Network.

The traditional traffic sign detection algorithm can not deal with the application scenarios such as intelligent transportation system or advanced assisted driving environment, and it is difficult to meet the application requirements in detection accuracy and efficiency. Focusing on the above problems, this paper proposes a traffic sign detection algorithm based on Single Shot Multibox Detector (SSD) combined with Receptive Field Module (RFM) and Path Aggregation Network (PAN). The proposed algorithm is abbreviated to SSD-RP. The SSD-RP uses the RFM to improve the receptive field and semantics of the predicted feature maps, thus improving the detection performance of small traffic signs. At the same time, the path aggregation network is introduced to integrate multiscale features, which makes the abstract semantic information and rich detailed information shared among multiscale feature maps, enhances the discrimination ability of feature system, and improves the location and classification accuracy of traffic signs. Following that, the spatial pyramid pooling module is used to pool the shallow features and integrate them into the bottom-up information transmission path of the path aggregation network, thus continuing to supplement the fine-grained features for the feature system and further improve the detection performance. The experimental results on GTSDB and CCTSDB data sets show that SSD-RP has higher mean average precision (mAP) than traditional SSD algorithm and can better detect small traffic signs, which means that SSD-RP has higher detection precision. In addition, the experimental results also show that, compared with the common object detection algorithms such as Faster R-CNN, RetinaNet, and YOLOv3, the SSD-RP achieves a better balance between detection time and detection precision.

An English Teaching Pronunciation Detection and Recognition Algorithm Based on Cluster Analysis and Improved SSD

The accuracy of English pronunciation is the key index to evaluate the quality of English teaching. Correct pronunciation and smooth language flow are the expectations of every student for English learning. Aiming at the poor effect and slow speed of the original SSD (Single Shot MultiBox Detector) algorithm in English teaching pronunciation detection, this paper proposes a clustering and improved SSD algorithm for English teaching pronunciation detection and recognition. The algorithm improves the concept module to enhance the feature extraction ability of the network and improve the detection speed. Meanwhile, it integrates multiscale features to realize multilayer multiplexing and equalization of features, so as to improve the detection effect of small target sound. This algorithm extracts more features by introducing channel attention mechanism, which increases the detection accuracy while reducing computation. In order to optimize the network’s ability to extract target location information, K-means clustering method is used to set the default parameters that are more in line with the characteristics of target samples. The experimental results showed that the proposed algorithm can accurately evaluate the pronunciation quality of reading aloud, so as to effectively reflect the oral English level of the reader.

UAV Image Small Object Detection Based on Composite Backbone Network

Small objects in traffic scenes are difficult to detect. To improve the accuracy of small object detection using images taken by unmanned aerial vehicles (UAV), this study proposes a feature-enhancement detection algorithm based on a single shot multibox detector (SSD), named composite backbone single shot multibox detector (CBSSD), which uses a composite connection backbone to enhance feature representation. First, to enhance the detection effect of small objects, the lead backbone network, VGG16, is kept constant, and ResNet50 is added as an assistant backbone network, and the residual structure in ResNet50 is used to obtain lower feature information. The obtained lower feature information is then fused to the lead network through feature fusion, allowing the lead network to retain rich lower feature information. Finally, the lower feature information in the prediction layer increases. The experimental results show that CBSSD has a significantly higher recognition rate and a lower false detection rate than conventional algorithms, and it still maintains a good detection effect under low illumination. This is of great significance to small object detection using images taken by UAVs in traffic scenes. Furthermore, a method to improve the SSD algorithm is proposed.

Individual automatic detection and identification of big cats with the combination of different body parts.

The development of facial recognition technology has become an increasingly powerful tool in wild animal individual recognition. In this paper, we develop an automatic detection and recognition method with the combinations of body features of big cats based on the deep convolutional neural network (CNN). We collected dataset including 12244 images from 47 individual Amur tigers (Panthera tigris altaica) at the Siberian Tiger Park by mobile phones and digital camera and 1940 images and videos of 12 individual wild Amur leopard (Panthera pardus orientalis) by infrared cameras. First, the single shot multibox detector algorithm is used to perform the automatic detection process of feature regions in each image. For the different feature regions of the image, like face stripe or spots, CNNs and multi-layer perceptron models were applied to automatically identify tiger and leopard individuals, independently. Our results show that the identification accuracy of Amur tiger can reach up to 93.27% for face front, 93.33% for right body stripe, and 93.46% for left body stripe. Furthermore, the combination of right face, left body stripe, and right body stripe achieves the highest accuracy rate, up to 95.55%. Consequently, the combination of different body parts can improve the individual identification accuracy. However, it is not the higher the number of body parts, the higher the accuracy rate. The combination model with 3 body parts has the highest accuracy. The identification accuracy of Amur leopard can reach up to 86.90% for face front, 89.13% for left body spots, and 88.33% for right body spots. The accuracy of different body parts combination is lower than the independent part. For wild Amur leopard, the combination of face with body spot part is not helpful for the improvement of identification accuracy. The most effective identification part is still the independent left or right body spot part. It can be applied in long-term monitoring of big cats, including big data analysis for animal behavior, and be helpful for the individual identification of other wildlife species.

Research on face tracking Algorithm Based on Detection and Supervision Tracking

Accurate face detection and tracking is widely used in many social life scenes. However, the uncontrollable background noise and random illumination change in the application scene will reduce the detection accuracy of the tracked target, and the rotation, occlusion and overlap of the tracked target will also affect the accuracy and success rate of the tracking algorithm. In order to solve the above problems, this paper proposes a method that uses deep learning detection method to supervise the correlation filter tracking algorithm to improve the success rate of de-tection and tracking. First of all, in the first frame of the picture, we use the deep learning SSD (Single Shot MultiBox Detector) algorithm to detect the face, and take the detected face as the tracking target, and use the correlation filtering algorithm DSST (Discriminative Scale Space Tracker) Tracking in the process of tracking, face detection is continuously carried out on the tracking target, and the detection results are used to monitor the tracking results, so as to reduce the target drift caused by the boundary effect, thus improving the accuracy of the tracking algorithm. The algorithm is tested and verified on OTB100 data set. The final experimental results show that the accuracy of this algorithm is obviously better than the mainstream classical algorithm, and the frame rate meets the real-time re-quirements.

Automatic Detection of Medial and Lateral Compartments from Histological Sections of Mouse Knee Joints Using the Single-Shot Multibox Detector Algorithm.

ObjectiveAlthough mouse osteoarthritis (OA) models are widely used, their histological analysis may be susceptible to arbitrariness and inter-examiner variability in conventional methods. Therefore, a method for the unbiased scoring of OA histology is needed. In this study, as the first step for establishing this system, we developed a computer-vision algorithm that automatically detects the medial and lateral compartments of mouse knee sections in a rigorous and unbiased manner.DesignA total of 706 images of coronal sections of mouse knee joints stained by hematoxylin and eosin, safranin O, or toluidine blue were randomly divided into training and validation images at a ratio of 80:20. A model to detect both compartments automatically was built by machine learning using a single-shot multibox detector (SSD) algorithm with training images. The model was tested to determine whether it could accurately detect both compartments by analyzing the validation images and 52 images of sections stained with Picrosirius red, a method not used for the training images.ResultsThe trained model accurately detected both medial and lateral compartments of all 140 validation images regardless of the staining method employed, severity of articular cartilage defects, and the anatomical positions and conditions of the sections. Our model also correctly detected both compartments of 50 of 52 Picrosirius red–stained images.ConclusionsBy applying deep learning based on the SSD algorithm, we successfully developed a model that detects the locations of the medial and lateral compartments of tissue sections of mouse knee joints with high accuracy.

Water surface object detection using panoramic vision based on improved single-shot multibox detector

In view of the deficiencies in traditional visual water surface object detection, such as the existence of non-detection zones, failure to acquire global information, and deficiencies in a single-shot multibox detector (SSD) object detection algorithm such as remote detection and low detection precision of small objects, this study proposes a water surface object detection algorithm from panoramic vision based on an improved SSD. We reconstruct the backbone network for the SSD algorithm, replace VVG16 with a ResNet-50 network, and add five layers of feature extraction. More abundant semantic information of the shallow feature graph is obtained through a feature pyramid network structure with deconvolution. An experiment is conducted by building a water surface object dataset. Results showed the mean Average Precision (mAP) of the improved algorithm are increased by 4.03%, compared with the existing SSD detecting Algorithm. Improved algorithm can effectively improve the overall detection precision of water surface objects and enhance the detection effect of remote objects.

Surface Detection of Solid Wood Defects Based on SSD Improved with ResNet

Due to the lack of forest resources in China and the low detection efficiency of wood surface defects, the output of solid wood panels is not high. Therefore, this paper proposes a method for detecting surface defects of solid wood panels based on a Single Shot MultiBox Detector algorithm (SSD) to detect typical wood surface defects. The wood panel images are acquired by an independently designed image acquisition system. The SSD model included the first five layers of the VGG16 network, the SSD feature mapping layer, the feature detection layer, and the Non-Maximum Suppression (NMS) module. We used TensorFlow to train the network and further improved it on the basis of the SSD network structure. As the basic network part of the improved SSD model, the deep residual network (ResNet) replaced the VGG network part of the original SSD network to optimize the input features of the regression and classification tasks of the predicted bounding box. The solid wood panels selected in this paper are Chinese fir and pine. The defects include live knots, dead knots, decay, mildew, cracks, and pinholes. A total of more than 5000 samples were collected, and the data set was expanded to 100,000 through data enhancement methods. After using the improved SSD model, the average detection accuracy of the defects we obtained was 89.7%, and the average detection time was 90 ms. Both the detection accuracy and the detection speed were improved.

Detection and Avoidance of Barriers in Driverless Cars using Deep Learning Techniques

In the advance of the technology and implantation of Internet of Things (IoT), the realization of smart city seems to be very needed. One of the key parts of a cyber-physical system of urban life is transportation. Such mission-critical application has led to inquisitiveness in researchers to develop autonomous robots from academicians and industry. In the domain of autonomous robot, intelligent video analytics is very crucial. By the advent of deep learning many neural ¬¬¬networks-based learning approaches are considered. One of the advanced Single Shot Multibox Detector (SSD) method is exploited for real-time video/image analysis using an IOT device and vehicles/any barrier avoidance on road is done using image processing. The proposed work makes use of SSD algorithm which is responsible for object detection and image processing to control the car, based on its current input. Thus, this work aims to develop real-time barrier detection and barrier avoidance for autonomous robots using a camera and barrier avoidance sensor in an unstructured environment.

Improved single shot multibox detector target detection method based on deep feature fusion

SummaryThe feature layers of different layers in the single shot multibox detector (SSD) are independently used as the input of the classification network, so it is easy to detect the same object. This article proposes an improved SSD model based on deep feature fusion. In the SSD algorithm, the deep feature fusion between the target detection layer and its adjacent feature layer is used, including convolution kernels and pooling kernels of different sizes, down‐sampling of low‐level features and up‐sampling of deconvolution of high‐level features. The network is improved by combining the target frame recommendation strategy in the SSD algorithm and the frame regression algorithm. The experimental results show that the improved SSD algorithm improves the detection accuracy and detection rate of the target, and the effect is more obvious for the relatively small‐scale target.

An intelligent water level monitoring method based on SSD algorithm

Water levels are essential components for the observation and management of water resources. However, the existing water level monitoring methods either require manpower, which is inefficient, or are restricted to a certain environment. This paper proposes a novel approach that can automatically monitor, recognize and calculate the water level based on deep learning. First, a series of experiments were performed in a physical pool to obtain images from real scenes. Then, the original Single Shot MultiBox Detector (SSD) preprocessing model was trained and optimized. Subsequently, a trained and verified staff gauge detection model was obtained, which was applied to extract the staff gauge information from images. Finally, a 24-hour time series of water level changes was simulated and analyzed using photographic images principle, and the NSE and R2 were 0.98 and 0.99, respectively. These results indicate that the proposed method is more accurate in practice and can provide a relatively accurate and reliable measurement technique.

Research on workpiece location algorithm based on improved SSD

PurposeTo improve production efficiency, industrial robots are expected to replace humans to complete the traditional manual operation on grasping, sorting and assembling workpieces. These implementations are closely related to the accuracy of workpiece location. However, workpiece location methods based on conventional machine vision are sensitive to the factors such as light intensity and surface roughness. To enhance the robustness of the workpiece location method and improve the location accuracy, a workpiece location algorithm based on improved Single Shot MultiBox Detector (SSD) is proposed.Design/methodology/approachThe proposed algorithm integrates a weighted bi-directional feature pyramid network into SSD. A feature fusion architecture is structured by the combination of low-resolution, strong semantic features and high-resolution, weak semantic features. The architecture is built through a top-down pathway, bottom-up pathway, lateral connections and skip connections. To avoid treating all features equally, learnable weights are introduced into each feature layer to characterize its importance. More detailed information from the low-level layers is injected into the high-level layers, which could improve the accuracy of workpiece location.FindingsIt is found that the maximum location error at the center point calculated from the proposed algorithm is decreased by more than 22% compared with that of the SSD algorithm. Besides, the average location error evolves a decrease by at least 5%. In the trajectory prediction experiment of the workpiece center point, the results of the proposed algorithm demonstrate that the average location error is below 0.13 mm and the maximum error is no more than 0.23 mm.Originality/valueIn this work, a workpiece location algorithm based on improved SSD is developed to extract the center point of the workpiece. The results demonstrate that the proposed algorithm is beneficial for workpiece location. The proposed algorithm can be readily used in a variety of workpieces or adapted to other similar tasks.

Application of portrait recognition system for emergency evacuation in mass emergencies

Abstract A portrait recognition system can play an important role in emergency evacuation in mass emergencies. This paper designed a portrait recognition system, analyzed the overall structure of the system and the method of image preprocessing, and used the Single Shot MultiBox Detector (SSD) algorithm for portrait detection. It also designed an improved algorithm combining principal component analysis (PCA) with linear discriminant analysis (LDA) for portrait recognition and tested the system by applying it in a shopping mall to collect and monitor the portrait and establish a data set. The results showed that the missing detection rate and false detection rate of the SSD algorithm were 0.78 and 2.89%, respectively, which were lower than those of the AdaBoost algorithm. Comparisons with PCA, LDA, and PCA + LDA algorithms demonstrated that the recognition rate of the improved PCA + LDA algorithm was the highest, which was 95.8%, the area under the receiver operating characteristic curve was the largest, and the recognition time was the shortest, which was 465 ms. The experimental results show that the improved PCA + LDA algorithm is reliable in portrait recognition and can be used for emergency evacuation in mass emergencies.

Survey of Target Detection Based on Neural Network

Target detection is a hot topic in the field of artificial intelligence, which is widely used in robot, UAV, aerospace and other fields. In this paper, the research background and significance of target detection are summarized, and two categories of target detection algorithms based on deep learning, i.e., candidate region based and regression based, are described. For the first category, a series of region with convolutional neural network (r-cnn) algorithms are introduced, this paper introduces the researchers’ research on the basis of r-cnn algorithm: the improvement of feature extraction network, pooling layer of region of interest and non-maximum suppression algorithm. The second algorithm is divided into Yolo (you only look once) series, SSD (single shot multibox detector) algorithm and its improvement. According to the current target detection algorithm in the development of more efficient and reasonable development trend, the research hotspot of target detection algorithm in the future is prospected, including unsupervised and unknown class object detection.

Research on fatigue detection based on visual features

The high incidence of traffic accidents brings immeasurable losses to life. In order to avoid such crises, researchers and automakers have used many methods to solve this problem. Among them, technology based on visual features is widely used in driver fatigue detection. As fatigue detection plays a vital role in the driving process, the high accuracy of fatigue monitoring is very important. This paper focuses on the method based on convolutional neural network to detect driver fatigue. First, in the face detection part, the Single-Shot Multi-Box Detector algorithm is used to improve the speed and accuracy of face detection to extract the eye and mouth regions; second, the VGG16 network is used to learn fatigue features, which is performed on the NTHU-Drowsy Driver Detection (NTHU-DDD) data set and the other two modified data sets Training test. The main result of this work is that the accuracy of fatigue monitoring is higher than other methods including the original method, with an accuracy rate of over 90%. And it has better generalization ability than the multi-physical feature fusion detection method. At the same time, we propose the fatigue detection method based on convolutional neural network to improve the advanced driver assistance system (ADAS) to make it more robust and reliable decision making.

Improved SSD-assisted algorithm for surface defect detection of electromagnetic luminescence

Defect detection of electromagnetic luminescence (EL) cells is the core step in the production and preparation of solar cell modules to ensure conversion efficiency and long service life of batteries. However, due to the lack of feature extraction capability for small feature defects, the traditional single shot multibox detector (SSD) algorithm performs not well in EL defect detection with high accuracy. Consequently, an improved SSD algorithm with modification in feature fusion in the framework of deep learning is proposed to improve the recognition rate of EL multi-class defects. A dataset containing images with four different types of defects through rotation, denoising, and binarization is established for the EL. The proposed algorithm can greatly improve the detection accuracy of the small-scale defect with the idea of feature pyramid networks. An experimental study on the detection of the EL defects shows the effectiveness of the proposed algorithm. Moreover, a comparison study shows the proposed method outperforms other traditional detection methods, such as the SIFT, Faster R-CNN, and YOLOv3, in detecting the EL defect.

A Vehicular Queue Length Measurement System in Real-Time Based on SSD Network

Abstract Vehicular queue length measurement is an important parameter to detect the traffic congestion, which is resulted from several issues such as traffic lights, accidents, and poor roads infrastructures. In this paper, a system in real-time is proposed to detect and measure the vehicular queue length at intersections. The proposed system consists of two main steps: the first step is the detection of queue by using frames differencing method to detect the motion in the target areas. If there is no a motion, then the second step is implemented to detect the vehicles in these areas by using Single Shot Multibox Detector (SSD) algorithm. If there are vehicles, that means the queue exists and the measurement process begins. Some modifications are applied on SSD algorithm to fit with in our system and to improve the accuracy of the vehicle detection process. The system is applied on videos obtained by stationary cameras. The experiments demonstrate that this system is able to accurately detect and measure the vehicular queue length.

Recognizing human behaviors from surveillance videos using the SSD algorithm

The aim is to better recognize human behaviors from surveillance videos. Human behavior recognition technology based on surveillance videos is researched, given the intellectual development of massive surveillance video data with full coverage. This technology builds a human behavior detection and recognition model using the new Single Shot MultiBox Detector (SSD) algorithm, which improves the recognition accuracy. The constructed model’s effectiveness is verified through comparisons with other traditional human behavior recognition algorithms via the TensorFlow framework. Results demonstrate the SSD model-based recognition algorithm’s accuracy is significantly higher than that of Direct Part Marking and Fast Convolutional Neural Network (CNN) algorithms. SSD’s average speed is 0.146 s/frame, and the average accuracy on different datasets is 82.8%. If the target is close or partially occluded, the SSD algorithm can also accurately detect the central target, and the detection efficiency is twice that of the R-CNN algorithm. The algorithm proposed has a simple structure and fast processing speed, which can solve the problems in target detection. The research results can provide a theoretical basis for the research on target detection related to human behavior recognition.

Small Object Recognition Algorithm of Grain Pests Based on SSD Feature Fusion

The detection of grain pests is of great significance to grain storage. However, in practice, because the size of grain insects is too small to identify. In this paper, the feature fusion SSD(single shot multi-box detector) algorithm based on the Top-Down strategy was proposed. Firstly, the Top-Down module is used to fuse the output characteristics of conv4 and conv5, and the block 11 which is not conducive to small object detection is deleted. Secondly, K-means clustering algorithm is used to cluster prior bounding boxes and made them more suitable for grain pests, which improves the performance on small object detection of grain pests. Five methods were used to enhance the self-made dataset of grain pests, and the enhanced dataset reached 9990 images. Experiments on the enhanced dataset show that the optimized model achieves a mAP (mean Average Precision) 96.89% with detection speed of 0.040s per image. Compared with 95.45% of the mAP achieved by the original SSD algorithm, the proposed model has a great improvement on detection performance. Compared with two-stages Faster R-CNN (mAP is 90.53% and speed is 0.115s per image), YOLOv3, TDFSSD and EfficentDet (D2, D1), the speed and accuracy of the optimized SSD algorithm have obvious advantages. The experimental results show that the proposed SSD model has a good performance on small object detection of grain pests and has a certain guiding significance for subsequent grain pest image detection.

The Assisted Positioning Technology for High Speed Train Based on Deep Learning

In the positioning process of a high-speed train, cumulative error may result in a reduction in the positioning accuracy. The assisted positioning technology based on kilometer posts can be used as an effective method to correct the cumulative error. However, the traditional detection method of kilometer posts is time-consuming and complex, which greatly affects the correction efficiency. Therefore, in this paper, a kilometer post detection model based on deep learning is proposed. Firstly, the Deep Convolutional Generative Adversarial Networks (DCGAN) algorithm is introduced to construct an effective kilometer post data set. This greatly reduces the cost of real data acquisition and provides a prerequisite for the construction of the detection model. Then, by using the existing optimization as a reference and further simplifying the design of the Single Shot multibox Detector (SSD) model according to the specific application scenario of this paper, the kilometer post detection model based on an improved SSD algorithm is established. Finally, from the analysis of the experimental results, we know that the detection model established in this paper ensures both detection accuracy and efficiency. The accuracy of our model reached 98.92%, while the detection time was only 35.43 ms. Thus, our model realizes the rapid and accurate detection of kilometer posts and improves the assisted positioning technology based on kilometer posts by optimizing the detection method.