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

Vehicle classifier training method

A training method and classifier technology, applied in the field of vehicle image recognition and detection, can solve the problems of long training time, paralysis of the classifier training process, insufficient memory, etc., and achieve the effect of optimizing the training process

Inactive Publication Date: 2017-02-01
GUANGDONG UNIV OF TECH
View PDF8 Cites 11 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The classifiers in the prior art use the above technical solutions for training and learning, which may cause insufficient memory and long training time as the number of detection samples increases, and even cause the training process of the classifier to be paralyzed.

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
  • Vehicle classifier training method
  • Vehicle classifier training method
  • Vehicle classifier training method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0030] please see Figure 1 to Figure 4 In this embodiment, 2000 vehicle positive sample pictures and 5000 vehicle negative sample pictures are manually selected, and the vehicle positive sample pictures and vehicle negative sample pictures are normalized to a picture of 64×64 pixels, wherein, the vehicle positive sample picture The selection standard is to include the left and right sides of the vehicle and expand 10% left and right, and include all images under the front car and expand 10% downward; the vehicle negative sample image is a road or natural scene image without a vehicle.

[0031] After normalizing the positive and negative vehicle images, the integral channel features are used to characterize the positive and negative images to form feature vectors.

[0032] The integral channel feature trains the positive and negative sample pictures with a picture size of 64×64 pixels, and speeds up the training by reducing the trained picture by a certain number of times. The...

Embodiment 2

[0065] This embodiment is the same as Embodiment 1 except for the following features: After normalizing the vehicle positive sample picture and the vehicle negative sample picture, HOG features are used to characterize the normalized positive and negative sample pictures to form features.

[0066] In this characterization process, the number of occurrences of local orientation gradients in positive and negative sample images will be counted. In the positive and negative sample images, the appearance and shape of local objects can be well described by the direction density distribution of gradients or edges. Therefore, the positive and negative sample pictures are divided into small connected regions (cell units), and then the gradient or edge direction histogram of each pixel in the cell unit is collected, and finally these histograms are combined to form a feature descriptor. The steps are as follows:

[0067] 1) Preprocess the positive and negative sample images, and then c...

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 vehicle classifier training method, which comprises the following steps: utilizing a few samples to train an initial classifier; converting videos collected by a camera sensor into pictures; carrying out detection on the pictures through the initial classifier; then, keeping the pictures passing detection of the initial classifier and pictures comprising a vehicle target but failing the detection of the classifier, wherein the pictures passing detection of the initial classifier may comprise correct detected vehicle pictures and wrong detected vehicle pictures; defining the wrong detected vehicle pictures as a negative sample hard example, and defining the pictures comprising the vehicle target but failing the detection of the classifier as a positive sample hard example; and adding the positive sample hard example and the negative sample hard example to a positive and negative sample library and continuing training a vehicle classifier until detection rate and false detection rate of the trained classifier reach requirements of clients. The method can increase diversity of classifier training samples, improve the detection rate of vehicle detection and reduce the false detection rate, and finally, optimizes the classifier training flow.

Description

technical field [0001] The invention relates to the technical field of vehicle image recognition and detection, in particular to a training method for a vehicle classifier. Background technique [0002] At this stage, vehicle image recognition and detection technology has been applied and paid attention to in the field of automobile assisted safe driving, and improving the detection rate and reducing the false detection rate are key indicators in the application of vehicle image recognition and detection technology. [0003] In the prior art, the vehicle image recognition and detection technology uses classifiers to learn to detect the representation of the target, and most of the classifiers are trained based on HaaR, HOG, and LBP features, or use Adaboost learning algorithm, or SVM learning algorithm trained. The classifiers in the prior art use the above technical solutions for training and learning, which may easily lead to insufficient memory, long training time, and e...

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): G06K9/62
CPCG06F18/217G06F18/24G06F18/214
Inventor 谢晶梅
Owner GUANGDONG UNIV OF TECH
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