Unlock instant, AI-driven research and patent intelligence for your innovation.

Efficient real-time target detection method applied to edge device

A target detection and edge device technology, applied in the field of computer vision, can solve the problems of missing low-level features and the limitation of the number of output scales, and achieve the effect of smooth loss curve, low loss value, and reasonable and reliable results

Pending Publication Date: 2022-07-29
SOUTHWEST PETROLEUM UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Furthermore, existing feature interaction methods are limited by the number of output scales, thus missing important low-level features

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Efficient real-time target detection method applied to edge device
  • Efficient real-time target detection method applied to edge device
  • Efficient real-time target detection method applied to edge device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0117] The data used in this example comes from the public data set BDD100K (Fisher Yu et al., 2020), and the pictures used are all from the pictures taken by the on-board camera, using image scaling and rotation transformation:

[0118]

[0119] like figure 1 shown.

[0120] The processed image is input to the convolution layer to extract the information in the input image, which is called image features such as figure 2 As shown, these features are represented by each pixel in the image in a combined or independent way, such as the texture feature of the image, the color feature:

[0121] N=(W-F+2P)÷(S+1)

[0122] After the nonlinear activation function ELU activation function, residual block and MHSA multi-head attention module:

[0123]

[0124] z [l+2] =W [l+2] a [l+1] +b [l+2 ]

[0125] Attention(q,k,v,r)=softmax(qk T +qr T )v

[0126] The structure diagram of MHSA multi-head attention module is as follows image 3 shown.

[0127] The extracted featur...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

According to the efficient real-time target detection method applied to the edge device provided by the invention, on the basis of a single-stage anchor-frame-free target detection algorithm, a brand-new feature extraction backbone network, a brand-new feature fusion structure and a brand-new detection head are subjected to experimental verification; the problems that feature extraction is insufficient, feature fusion is incomplete, information interaction is poor and the effect after deployment is not ideal are solved, the two globally recognized data sets including the COCO data set and the VOC data set are combined, the globally optimal training model is selected in the process of training the model and debugging the model, the obtained model parameters can make classification more accurate and positioning more accurate, and the method is suitable for large-scale popularization and application. And meanwhile, the size of the model is more suitable for being applied to edge equipment with limited hardware conditions, so that the obtained target detection model better meets the requirement of real-time target detection, and the obtained result is more reasonable and reliable. The method is not only used for target detection, but also can be popularized and used in the fields of segmentation, face recognition and the like, and has wide application value.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an efficient real-time target detection method applied to edge devices. Background technique [0002] Object detection is one of the most fundamental problems in computer vision, which locates and classifies various objects in images or videos. Due to its fundamental nature, object detection has numerous applications in various fields such as unmanned vehicles (UAVs), medical imaging, identity verification, robot navigation, motion analysis, and more. etc., and has been adapted for more complex problem statements such as object tracking, action recognition, facial recognition, etc. Many of these applications require highly accurate real-time feedback. In recent years, object detection has achieved significant progress in both accuracy and efficiency due to the use of deep learning algorithms. Object detection algorithms are usually divided into two types: one is two-st...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06V10/44G06V10/26G06V10/764G06V10/82G06N3/08G06N3/04G06K9/62
CPCG06V10/44G06V10/26G06V10/82G06V10/764G06N3/08G06N3/045G06F18/2415
Inventor 程茂凯陈金令徐紫涵
Owner SOUTHWEST PETROLEUM UNIV