CN105608444B - A kind of wild animal image-recognizing method for automatic Pilot - Google Patents
A kind of wild animal image-recognizing method for automatic Pilot Download PDFInfo
- Publication number
- CN105608444B CN105608444B CN201610055926.XA CN201610055926A CN105608444B CN 105608444 B CN105608444 B CN 105608444B CN 201610055926 A CN201610055926 A CN 201610055926A CN 105608444 B CN105608444 B CN 105608444B
- Authority
- CN
- China
- Prior art keywords
- image
- wild animal
- neural network
- matrix
- center
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
Abstract
The present invention relates to a kind of wild animal image-recognizing methods for automatic Pilot, comprising the following steps: carries out manifold learning to acquisition image using LLE algorithm, obtains characteristics of image；Neural network is trained by characteristics of image；The image acquired in real time substitution neural network is identified, obtains wild animal type label, and the corresponding wild animal type of the label is notified into driver.The present invention has incorporated the thought of manifold learning, and the low-dimensional by obtaining data is expressed, to preferably represent the original feature of data.The hidden node of neural network, and the initial center of the c class due to setting are sought using K-means algorithm, the number of iterations of K-means algorithm is greatly decreased, reduces algorithm time loss, and arithmetic speed is fast.
Description
Technical field
The present invention relates to wild animal image-recognizing method, specifically a kind of wild animal image for automatic Pilot
Recognition methods.
Background technique
With the high speed development of human civilization, the coverage of mankind's activity also is accelerating to expand, continuous with highway
It builds, more and more highways have been built in the scope of activities of wild animal, and many wild animals are faced with survival pressure just
It is to fight for living space with the mankind, the conflict of this type is often seen in our life.It is wild such as on certain highway
Lively object jaywalks suddenly, the driver of automobile in the process of moving it is not expected that, cause unnecessary injury.It is no matter right
In the mankind and the earth be all a kind of loss.
As automotive performance is promoted, automobile comes the scope of activities of the mankind increasingly as a kind of tool ridden instead of walk,
Even threatened the daily lifes of animals, and the high speed of automobile is quite dangerous for animals, animal without
Method effectively hides the automobile for telling highway, how to be always one to driver's appearance of wild animal on road that steps up vigilance
Problem.
Existing technology is mostly for the detection of pedestrian, and for the detection of wild animal, there is presently no and animal and people
Difference be the diversity of animal, and manifold learning arithmetic is rarely used image procossing skill to the extraction effect of feature
Art.
Summary of the invention
Insufficient in view of the above technology, it is an object of the invention to provide a kind of wild animal image recognition sides for automatic Pilot
Method.
The technical solution adopted by the present invention to solve the technical problems is: a kind of wild animal image for automatic Pilot
Recognition methods, comprising the following steps:
Manifold learning is carried out to acquisition image using LLE algorithm, obtains characteristics of image；
Neural network is trained by characteristics of image；
The image acquired in real time substitution neural network is identified, obtains wild animal type label, and by the label
Corresponding wild animal type notifies driver.
It is described using LLE algorithm to acquisition image carry out manifold learning the following steps are included:
Schemed using image as sample architecture k- neighbour, and calculates the similarity between any two image as approximate geodetic
Linear distance:
min(dG(i,j),dG(i,k)+dG(k,j))
Wherein, dGFor the Euclidean distance between any two image on k- neighbour's figure, image index i, j, k 1,
2 ..., N, wherein N is image number；
Structural matrix M=(I-W)T(I-W), wherein I is N × N unit matrix, and W is the nearly k- neighbour's figure matrix of N × N, i.e. k- is close
Neighbour schemes the approximate geodesic curve distance matrix between upper any two image；
Feature decomposition is carried out to Metzler matrix, X takes the preceding m feature vector of M as feature extraction as a result, i.e. characteristics of image
X1…Xm。
It is described neural network is trained by characteristics of image the following steps are included:
Using characteristics of image as input, type of vehicle label is output, and hidden node is what K-means algorithm clustered
Center；
Neural network is trained to obtain the weight of each node output of hidden layer.
The K-means algorithm the following steps are included:
1) select the initial center of c class: c is the one of several points of number of samples, and first sample is in data set
The heart, c-th of sample point farthest for c-1 data point before distance in all data points；Wherein data set is X, and data point indicates
Certain characteristics of image；
2) it to any one sample, asks it to arrive the distance at c center, which is grouped into where shortest center
Class；
3) point in each class is averaged to the cluster centre as such；Return step 2), until current all classes
Cluster centre and the last iteration cluster centre that obtains all classes difference be less than threshold value until.
It is described by the image acquired in real time substitution neural network identified the following steps are included:
The image acquired in real time is subjected to manifold learning to acquisition image using LLE algorithm, obtains characteristics of image；
Characteristics of image and weight are substituted into neural network, obtain wild animal type label:
YNFor type of vehicle label, W is weight, DNFor the distance matrix of current sample and each cluster centre, D is all poly-
The distance between class center matrix, p are hidden node number.
The invention has the following beneficial effects and advantage:
1. the present invention by the thought of manifold learning, obtains image in the expression of low-dimensional, to more effectively represent original
The feature of sample provides the obvious characteristic of sample for subsequent study and training.
2. the present invention has incorporated the thought of manifold learning, the low-dimensional by obtaining data is expressed, to preferably represent number
According to original feature.
It is initial 3. the present invention seeks the hidden node of neural network using K-means algorithm, and due to the c class of setting
The number of iterations of K-means algorithm is greatly decreased in center, reduces algorithm time loss, and arithmetic speed is fast.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is neural network schematic diagram of the invention.
Specific embodiment
The present invention will be further described in detail below with reference to the embodiments.
With the rise of manifold learning, high dimensional data then can preferably express the spy of higher-dimension in the potential manifold of low-dimensional
Sign, and the dimension that then can be seen that image pattern of the pixel of image can pass through the algorithm of manifold learning in this way and realize animal
The extraction of characteristics of image.
The present invention is expressed by the low-dimensional that manifold learning LLE algorithm obtains various wild animal images, as learning machine
Feature learns learning machine, and the purpose of the present invention makes running car to a certain region, and intelligent vehicle-mounted system can be automatic
Identify the wild animal for living in the region appeared on the road of front side, thus to driver with it is enough pay attention into
Row evacuation, specific steps are as follows.
As shown in Figure 1, specific steps are as follows:
1. for all known type wild animals, (the wild animal type of different zones is different, for different
The wild animal type of region detection is also different) feature extraction of image set, it will be wild based on all kinds of wild animals
Animal picture carries out feature extraction, is cut into the picture of fixed size, is schemed using image as sample architecture k- neighbour, that is, passes through meter
Calculate the similarity matrix that Euclidean distance calculates any two picture:
Picture is lined up into vector according to column, the distance between two vectors are constructed into neighbour as the similarity of image
Scheme, neighbour's number is the 5% of total number of samples in the present invention, i.e., it is the sample apart from the smallest preceding k for each sample
This neighbour.The preceding k width of every width picture and the picture are linked to be a line construction neighbour's figure, wherein have in neighbour's figure while while Wij
Value is 1, and it is 0 that remaining is boundless, calculates the approximate minimal geodesic distance between all the points:
min(dG(i,j),dG(i,k)+dG(k,j))
dGFor the Euclidean distance between any two image, i, j, k=1,2 ..., N, wherein N is image number；According to figure
The distance between midpoint is most short to calculate the distance between point, finally obtains the D distance matrix of sample data set, wherein i, j=
1,2 ..., N, wherein N is number of samples.
2. constructing M=(I-W)T(I-W), wherein I be N × N unit matrix, W be N × N neighbour figure matrix, to Metzler matrix into
Row feature decomposition, X take M before M feature vector as feature extraction as a result, what is obtained is then that the M dimension table of data set reaches.
3. k-means cluster is carried out to Y, first to all images as shown in Fig. 2, RBF neural network model is learnt
Obtained featureIt is clustered (X1 ... Xm in such as Fig. 2), M=m is input layer number, uses K-means
Algorithm is realized:
3-1 selects the initial center of c class: c is 1/10th of number of samples, first center for data set, the
Two points farthest for first point of distance in the data point of all compositions, third is in data set apart from the first two point distance
Farthest point ... and so on.Wherein data set is X, and data point indicates certain characteristics of image.
3-2 is in kth time iteration, to any one sample, asks it to arrive the distance at c center, which is grouped into distance
Class where shortest center；
3-3 is averaged using all points in such are calculated as updating such cluster centre；
3-4 is for c all cluster centres, if difference is less than after utilizing the iterative method of (3-2) (3-3) to update
0.01, then iteration terminates, and otherwise continues iteration.
Difference obtains the difference of the cluster centre of all classes for the cluster centre and last time iteration of current all classes.
Training sample: the mass center { C that will acquire1, C2,…,CqHidden node as neural network, i.e. in Fig. 2
Centered on centre indexing acquisition sample:
Distance DD of the calculating all the points to mass center；
Design factor σ
Wherein p is hidden layer number, and D is distance of all the points to mass center.
Calculate neuron input
By acquisitionVector adds coefficient of the independent a line that an element is 1 as RBF (radial base neural net)
Design factor matrix, that is, weight W
Wherein Y is the label of sample, is the Y in Fig. 2.
Also there is the output of response for the vehicle of the output of result other types being not concerned with, unified output is other classes
Type.
4. being predicted according to obtained parameter new samples
Calculate the Distance matrix D of target sample and mass center CN；
Calculate input coefficient
The column that added elements are 1 constitute RBF coefficient, and RBF coefficient obtains the tag along sort of target multiplied by weight
Finally obtained YNThe as label of new samples.
To the region of the road covering in image as target area during traveling, when camera is fixed
Selected, target area identified in real time, sounded an alarm if being found to be known sample type remind driver note that
Driver makes the further injury slowed down and can be then greatly reduced to wild animal again.
Claims (4)
1. a kind of wild animal image-recognizing method for automatic Pilot, it is characterised in that the following steps are included:
Manifold learning is carried out to acquisition image using LLE algorithm, obtains characteristics of image；
Neural network is trained by characteristics of image；
The image acquired in real time substitution neural network is identified, obtains wild animal type label, and the label is corresponding
Wild animal type notify driver；
It is described by the image acquired in real time substitution neural network identified the following steps are included:
The image acquired in real time is subjected to manifold learning to acquisition image using LLE algorithm, obtains characteristics of image；
Characteristics of image and weight are substituted into neural network, obtain wild animal type label:
YNFor wild animal type label, W is weight, DNFor the distance matrix of current sample and each cluster centre, D is all poly-
The distance between class center matrix, p are hidden node number.
2. a kind of wild animal image-recognizing method for automatic Pilot according to claim 1, it is characterised in that institute
State using LLE algorithm to acquisition image carry out manifold learning the following steps are included:
Schemed using image as sample architecture k- neighbour, and calculate the similarity between any two image as approximate geodesic curve away from
From:
min(dG(i,j),dG(i,k)+dG(k,j))
Wherein, dGFor the Euclidean distance between any two image on k- neighbour's figure, image index i, j, k 1,2 ..., N,
Middle N is image number；
Structural matrix M=(I-W)T(I-W), wherein I is N × N unit matrix, and W is N × N k- neighbour figure matrix, i.e. k- neighbour
Approximate geodesic curve distance matrix on figure between any two image；
Feature decomposition is carried out to Metzler matrix, X takes the preceding m feature vector of M as feature extraction as a result, i.e. characteristics of image X1 ...
Xm。
3. a kind of wild animal image-recognizing method for automatic Pilot according to claim 1, it is characterised in that institute
State neural network is trained by characteristics of image the following steps are included:
Using characteristics of image as input, wild animal type label is output, and hidden node is what K-means algorithm clustered
Center；
Neural network is trained to obtain the weight of each node output of hidden layer.
4. a kind of wild animal image-recognizing method for automatic Pilot according to claim 3, it is characterised in that institute
State K-means algorithm the following steps are included:
1) select the initial center of c class: c is the one of several points of number of samples, and first sample is the center of data set, the
The c sample point farthest for c-1 data point before distance in all data points；Wherein data set is X, and data point indicates certain figure
As feature；
2) it to any one sample, asks it to arrive the distance at c center, which is grouped into apart from the class where shortest center；
3) point in each class is averaged to the cluster centre as such；Return step 2), until gathering for current all classes
Until the difference for the cluster centre that class center and last iteration obtain all classes is less than threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610055926.XA CN105608444B (en) | 2016-01-27 | 2016-01-27 | A kind of wild animal image-recognizing method for automatic Pilot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610055926.XA CN105608444B (en) | 2016-01-27 | 2016-01-27 | A kind of wild animal image-recognizing method for automatic Pilot |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105608444A CN105608444A (en) | 2016-05-25 |
| CN105608444B true CN105608444B (en) | 2018-12-28 |
Family
ID=55988370
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610055926.XA Active CN105608444B (en) | 2016-01-27 | 2016-01-27 | A kind of wild animal image-recognizing method for automatic Pilot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105608444B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2599977A (en) * | 2020-04-28 | 2022-04-20 | Nvidia Corp | Notifications determined using one or more neural networks |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108074224B (en) * | 2016-11-09 | 2021-11-05 | 生态环境部环境规划院 | Method and device for monitoring terrestrial mammals and birds |
| CN109963460A (en) * | 2016-11-24 | 2019-07-02 | 夏普株式会社 | The control method and program of animal identifier, animal identifier |
| CN108171274B (en) * | 2018-01-17 | 2019-08-09 | 百度在线网络技术（北京）有限公司 | The method and apparatus of animal for identification |
| CN108460428A (en) * | 2018-04-11 | 2018-08-28 | 波奇（上海）信息科技有限公司 | A kind of method and apparatus of pet image recognition |
| CN108545021A (en) * | 2018-04-17 | 2018-09-18 | 济南浪潮高新科技投资发展有限公司 | A kind of auxiliary driving method and system of identification special objective |
| CN110824912B (en) * | 2018-08-08 | 2021-05-18 | 华为技术有限公司 | Method and apparatus for training a control strategy model for generating an autonomous driving strategy |
| CN109034109B (en) * | 2018-08-16 | 2021-03-23 | 新智数字科技有限公司 | Pedestrian re-identification method and device based on clustering algorithm |
| CN113397562A (en) * | 2021-07-20 | 2021-09-17 | 电子科技大学 | Sleep spindle wave detection method based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040017947A1 (en) * | 2002-07-29 | 2004-01-29 | Ming-Hsuan Yang | Extended Isomap using fisher linear discriminant and kernel fisher linear discriminant |
| CN101706871A (en) * | 2009-11-05 | 2010-05-12 | 上海交通大学 | Isometric mapping based facial image recognition method |
| CN101976373A (en) * | 2010-11-02 | 2011-02-16 | 上海电机学院 | Neural network structural design method based on high-dimensional space classifier |
| CN103020657A (en) * | 2012-12-28 | 2013-04-03 | 沈阳聚德视频技术有限公司 | License plate Chinese character recognition method |
-
2016
- 2016-01-27 CN CN201610055926.XA patent/CN105608444B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040017947A1 (en) * | 2002-07-29 | 2004-01-29 | Ming-Hsuan Yang | Extended Isomap using fisher linear discriminant and kernel fisher linear discriminant |
| CN101706871A (en) * | 2009-11-05 | 2010-05-12 | 上海交通大学 | Isometric mapping based facial image recognition method |
| CN101976373A (en) * | 2010-11-02 | 2011-02-16 | 上海电机学院 | Neural network structural design method based on high-dimensional space classifier |
| CN103020657A (en) * | 2012-12-28 | 2013-04-03 | 沈阳聚德视频技术有限公司 | License plate Chinese character recognition method |
Non-Patent Citations (1)
| Title |
|---|
| 基于LLE和BP神经网络的人脸识别;吴俊强 等;《激光杂志》;20060531;第27卷(第5期);第71-73页 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2599977A (en) * | 2020-04-28 | 2022-04-20 | Nvidia Corp | Notifications determined using one or more neural networks |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105608444A (en) | 2016-05-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105608444B (en) | A kind of wild animal image-recognizing method for automatic Pilot | |
| Dong et al. | A lightweight vehicles detection network model based on YOLOv5 | |
| Maheswari et al. | Intelligent fruit yield estimation for orchards using deep learning based semantic segmentation techniques—a review | |
| CN109117701B (en) | Pedestrian intention identification method based on graph convolution | |
| Brust et al. | Convolutional patch networks with spatial prior for road detection and urban scene understanding | |
| CN107609638B (en) | method for optimizing convolutional neural network based on linear encoder and interpolation sampling | |
| CN109961019A (en) | A kind of time-space behavior detection method | |
| CN111052151B (en) | Video action positioning based on attention suggestion | |
| CN108431826A (en) | Object in automatic detection video image | |
| CN104217214A (en) | Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method | |
| CN111898617B (en) | Target detection method and system based on attention mechanism and parallel void convolution network | |
| Wang et al. | A vehicle detection algorithm based on deep belief network | |
| CN104834748A (en) | Image retrieval method utilizing deep semantic to rank hash codes | |
| CN104281853A (en) | Behavior identification method based on 3D convolution neural network | |
| US20220301173A1 (en) | Method and system for graph-based panoptic segmentation | |
| CN107704924B (en) | Construction method of synchronous self-adaptive space-time feature expression learning model and related method | |
| WO2016175925A1 (en) | Incorporating top-down information in deep neural networks via the bias term | |
| Turay et al. | Toward performing image classification and object detection with convolutional neural networks in autonomous driving systems: A survey | |
| CN113269252A (en) | Power electronic circuit fault diagnosis method for optimizing ELM based on improved sparrow search algorithm | |
| JP6778842B2 (en) | Image processing methods and systems, storage media and computing devices | |
| De Menezes et al. | Object recognition using convolutional neural networks | |
| CN104021395B (en) | Target tracing algorithm based on high-order partial least square method | |
| Guo et al. | Underwater sea cucumber identification via deep residual networks | |
| Ilyas et al. | DAM: Hierarchical adaptive feature selection using convolution encoder decoder network for strawberry segmentation | |
| Yu et al. | TasselLFANet: a novel lightweight multi-branch feature aggregation neural network for high-throughput image-based maize tassels detection and counting |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |