Multi-block SSD based on small object detection for UAV railway scene surveillance

Li, Yundong; Han, Dong; Li, Hongguang; Zhang, Xueyan; Zhang, Baochang; Xiao, Zhifeng

doi:10.1016/j.cja.2020.02.024

Cited by 79 publications

(24 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…According to the characteristics of small object such as low resolution and complex background, the feature of small object in low-level network are more obvious. After several times of pooling of neural network, the feature maps of small object will appear fuzzy, causing bad detection results [26][27]. Therefore, we change the resolution of the input image from 300 × 300 to 428 × 428, and the network is optimized on SSD baseline.…”

Section: B Improved Ssd Network Structurementioning

confidence: 99%

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

et al. 2021

View full text Add to dashboard Cite

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.

show abstract

Section: B Improved Ssd Network Structurementioning

confidence: 99%

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

et al. 2021

View full text Add to dashboard Cite

show abstract

“…The YOLO algorithm has a weak generalization ability for object aspect ratio, and its detection accuracy decreases when the aspect ratio of a new class of objects appears [24][25][26]. SSD has two advantages, namely, real-time processing and high accuracy, especially when targeting objects in different size scales, meaning a certain level of accuracy can be guaranteed [27][28][29][30][31]. Given that the size of the ore hauling vehicle to be detected is generally variable and uncertain, the SSD algorithm is comparatively more suitable for the multi-scale detection of ore hauling vehicles.…”

Section: Introductionmentioning

confidence: 99%

Intelligent Localization Sampling System Based on Deep Learning and Image Processing Technology

Yang

Zhou

et al. 2022

Sensors

View full text Add to dashboard Cite

In this paper, deep learning and image processing technologies are combined, and an automatic sampling robot is proposed that can completely replace the manual method in the three-dimensional space when used for the autonomous location of sampling points. It can also achieve good localization accuracy, which solves the problems of the high labor intensity, low efficiency, and poor scientific accuracy of the manual sampling of mineral powder. To improve localization accuracy and eliminate non-linear image distortion due to wide-angle lenses, distortion correction was applied to the captured images. We solved the problem of low detection accuracy in some scenes of Single Shot MultiBox Detector (SSD) through data augmentation. A visual localization model has been established, and the image coordinates of the sampling point have been determined through color screening, image segmentation, and connected body feature screening, while coordinate conversion has been performed to complete the spatial localization of the sampling point, guiding the robot in performing accurate sampling. Field experiments were conducted to validate the intelligent sampling robot, which showed that the maximum visual positioning error of the robot is 36 mm in the x-direction and 24 mm in the y-direction, both of which meet the error range of less than or equal to 50 mm, and could meet the technical standards and requirements of industrial sampling localization accuracy.

show abstract

“…In this work a system using modified-SSD which is based on visual geometry group (VGG16) is used for human detection in surveillance cameras [9]. The CNN based network takes the input from CHOKEPOINT dataset, which has frames of a surveillance video.…”

Section: Introductionmentioning

confidence: 99%

Real-time human detection for electricity conservation using pruned-SSD and arduino

Ushasukhanya

Jothilakshmi

2021

IJECE

View full text Add to dashboard Cite

Electricity conservation techniques have gained more importance in recent years. Many smart techniques are invented to save electricity with the help of assisted devices like sensors. Though it saves electricity, it adds an additional sensor cost to the system. This work aims to develop a system that manages the electric power supply, only when it is actually needed i.e., the system enables the power supply when a human is present in the location and disables it otherwise. The system avoids any additional costs by using the closed circuit television, which is installed in most of the places for security reasons. Human detection is done by a Modified-single shot detection with a specific hyperparameter tuning method. Further the model is pruned to reduce the computational cost of the framework which in turn reduces the processing speed of the network drastically. The model yields the output to the Arduino micro-controller to enable the power supply in and around the location only when a human is detected and disables it when the human exits. The model is evaluated on CHOKEPOINT dataset and real-time video surveillance footage. Experimental results have shown an average accuracy of 85.82% with 2.1 seconds of processing time per frame.

show abstract

Multi-block SSD based on small object detection for UAV railway scene surveillance

Cited by 79 publications

References 11 publications

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

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

Intelligent Localization Sampling System Based on Deep Learning and Image Processing Technology

Real-time human detection for electricity conservation using pruned-SSD and arduino

Contact Info

Product

Resources

About