Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A target detection method in a vehicle-mounted environment

A target detection and environment technology, applied in the field of target detection in the vehicle environment

Pending Publication Date: 2019-05-10
SHENYANG JIANZHU UNIVERSITY +1
View PDF3 Cites 28 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, it is inevitable that there will be accidents caused by the driver's negligence to check the auxiliary system and the visual mirror

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
  • A target detection method in a vehicle-mounted environment
  • A target detection method in a vehicle-mounted environment
  • A target detection method in a vehicle-mounted environment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0113] Such as image 3 As shown, the present embodiment provides a target detection method in a vehicle environment, including:

[0114] S1. The on-vehicle processing device receives the video image frame collected by the image acquisition device on the vehicle;

[0115] S2. Perform preprocessing on each video image frame to obtain each preprocessed video image frame.

[0116] For example, normalize and grayscale each video image frame, and scale the grayscaled video image frame to a specification of 300*300*1 to obtain each preprocessed video image frame .

[0117] The effective detection distance of the tiny model in this embodiment is greater than 45m.

[0118] S3. Input each preprocessed video image frame to the pre-trained tiny model to obtain at least 8 feature maps of different scales corresponding to each frame;

[0119] S4. Select three feature maps from all feature maps in a multi-scale feature fusion method, and generate candidate area anchor boxes of different...

Embodiment 2

[0133] First, the dataset for training the tiny model

[0134] The training of the tiny model is mainly carried out on the PASCALVOC dataset and the self-collected dataset, and the training of the model adopts the strategy of transfer learning. Since images in a large truck in-vehicle environment are required for object detection, object detection models need to be trained on the same type of dataset. The images in the truck-mounted environment are similar to the public dataset PASCALVOC dataset, but the image perspective is different, and the amount of image data in the self-collected dataset is small, which is not enough to train the entire model from scratch. Migration learning refers to training a model on a public dataset, a process also known as pre-training. The model parameters are then fine-tuned on the self-collected dataset. This kind of transfer learning can make the model inherit the objectivity of the pre-trained model, and can train on less data to get better ...

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

The invention provides a target detection method in a vehicle-mounted environment. The method comprises the steps of a vehicle-mounted processing device receiving a video image frame; preprocessing each video image frame, and inputting the preprocessed video image frames into a pre-trained tiny model to obtain at least eight feature maps with different scales; For all the feature maps, generatingcandidate regions with different sizes and shapes; selecting candidate regions in all the feature maps for convolution operation to obtain four offset positions of the candidate regions for target positioning prediction and three category confidence coefficients for target classification prediction; Based on the offset position and the category confidence coefficient, performing prediction screening on each annotation frame, annotating the screened annotation frames on the video image frame, and annotating the predicted category; and S6, synthesizing the marked video image frame into a video,displaying the video on a display screen, and carrying out early warning prompt. According to the method, the target in the panoramic auxiliary system can be detected and marked with reminding.

Description

technical field [0001] The invention relates to target detection technology, in particular to a target detection method in a vehicle environment. Background technique [0002] With the continuous improvement of mining technology and the continuous increase of mining intensity, the large-scale equipment is also constantly developing. Large-scale truck transportation is one of the most important forms of production and transportation in large-scale mines. [0003] The characteristics of large trucks used in mine transportation are mainly as follows: the road space is complex, due to the certain mining range, it needs to continue to develop in depth, the transportation road conditions are complex, there are many curves and slopes, and the sight line is not good, and the road changes frequently , the working environment is harsh; [0004] There are many types of production auxiliary vehicles, such as command vehicles, gunpowder vehicles, bulldozers, sprinklers, etc.; truck dri...

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
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06N3/04
Inventor 王继春肖冬杨丰华
Owner SHENYANG JIANZHU UNIVERSITY
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com