CN105787456A - Method for detecting pedestrians in night far infrared images - Google Patents
Method for detecting pedestrians in night far infrared images Download PDFInfo
- Publication number
- CN105787456A CN105787456A CN201610119259.7A CN201610119259A CN105787456A CN 105787456 A CN105787456 A CN 105787456A CN 201610119259 A CN201610119259 A CN 201610119259A CN 105787456 A CN105787456 A CN 105787456A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- far infrared
- image
- night
- pedestrians
- 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.)
- Pending
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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Traffic Control Systems (AREA)
- Image Processing (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a method for detecting pedestrians in night far infrared images, and belongs to the field of image processing. The method comprises the steps of: 1, selecting a plurality of images of pedestrians and non-pedestrians in the night far infrared images, and respectively marking the images of the pedestrians and the non-pedestrians as positive and negative samples; 2, representing any positive and negative sample Si with a Haar characteristic, and generating a Haar characteristic vector Hi; 3, inputting the Haar characteristic vector Hi into a SVM classifier, training the SVM classifier and obtaining a vehicle classifier C; 4, reading an actual night far infrared image, and obtaining an area where the pedestrians may exist through a self-adaptive threshold T, wherein the area is defined as a pedestrian hypothesis; and 5; representing the pedestrian hypothesis to be judged with the Haar characteristic vector Hi, and inputting the Hi into the classifier C, wherein the type of the pedestrian hypothesis to be judged is output by the classifier C the pedestrian type or the non-pedestrian type. According to the invention, all pedestrians existing in the night far infrared image can be detected through verifying all pedestrian hypotheses, and the accurate detection of the pedestrians in the night far infrared image is realized.
Description
Technical field
The invention belongs to technical field of image processing, relate to image information perception, be specifically related to the pedestrian detection method in far infrared image at a kind of night.
Background technology
Traffic safety problem has become international significant problem, and the safety of automobile is self-evident especially on the impact of human life's property.Raising along with the development of highway and automotive performance, automobile driving speed is also accelerated accordingly, in addition the increase of automobile quantity and transportation are day by day busy, motor-vehicle accident increases caused casualties and property loss, having become a social problem that can not be ignored, the traffic safety of automobile more seems extremely important.Traditional passive security be far from avoided that traffic accident occur, and active safety technologies due to can trouble-saving generation and receive much attention.Visual sensing due to have contain much information, with low cost, have a wide range of applications in field of automotive active safety.
Pedestrian detection technology refers to and utilizes image sensing means the pedestrian in image is searched and judges, it is thus achieved that the process of many attribute (such as position, speed, shape, outward appearance) of pedestrian in image.It is field of automotive active safety, especially realizes one of key technology of anti-collision warning (CW) and automatic emergency brake (AEB) function.But, the visible detection method effect under night-environment adopting common CCD sensor at present is poor.This is because under night-environment, owing to lacking enough illumination, target imaging quality is low, cause traditional in a large number losing efficacy for the day by day vehicle detecting algorithm under good photoenvironment.Therefore, how to improve the accuracy of vehicle detection at night, robustness and real-time, solve the vehicle detection under real road environment at night and be still a content needing research.
Traditional camera is due to the restriction of image-forming principle, by illumination restriction substantially, it is difficult to be obviously improved nighttime imaging quality and investigative range.And the infrared ray that far infrared camera sends by detecting object carries out imaging, completely not by the illumination effect of visible waveband, its image still has good visuality at night.For this, the present invention is with far infrared imagery for means, it is proposed to the pedestrian detection method in a kind of night far infrared image.
Summary of the invention
For the problem in above-mentioned background technology, the present invention proposes the pedestrian detection method in far infrared image at a kind of night, for the imaging characteristics of vehicle in far infrared image, it is achieved a kind of effective vehicle checking method.Detailed technology scheme of the present invention is as follows:
Far infrared image capture device is utilized to obtain road at night time image;Utilize adaptive threshold fuzziness technology to obtain the region that in image, brightness is significantly greater, and produce pedestrian region that may be present hypothesis based on this;Adopt Haar feature that positive and negative training sample is characterized, then adopt support vector machine (SVM) as grader, Haar feature to be trained, obtain pedestrian detection grader;Utilizing this pedestrian detector to assume to be verified to region, the pedestrian in road image is detected by final realization.As it is shown in figure 1, comprise the following steps:
Step 1: by manually choosing the picture of pedestrian in far infrared image at a large amount of night and non-pedestrian, be respectively labeled as positive sample and negative sample.All positive samples and negative sample are all normalized to length and width and is the size of 32x16 pixel.If positive sample and negative sample number are altogether for n (generally requiring that positive number of samples is more than 5000, negative sample number is more than 20000).
Step 2: by any one positive negative sample SiCharacterize by Haar feature, generate Haar characteristic vector Hi, (i=1,2 ..., n).
Step 2-1: choose 10 Like-Fenton Oxidation as shown in Figure 2.
Step 2-2: the Haar feature calculation numerical value of dissimilar, yardstick and position just constitutes a multidimensional Haar characteristic series vector H of certain samplei, (i=1,2 ..., n).
Step 3: utilize grader to be trained.The Haar characteristic vector H that will generate in step 2iIt is input in SVM classifier and is trained, can train and obtain vehicle classification device C.
Step 4: read in far infrared image Img at actual night, and obtain pedestrian by adaptive threshold T and would be likely to occur region.In far infrared image under night-environment, part thermal objects is (such as positions such as wheel of vehicle, electromotor front decks, also pedestrian is included) obvious high brightness is presented because temperature is higher, and in road, most object such as the road surface in image, number presents low-light level because temperature is low.Therefore, a suitable threshold value T can be passed through to be split.The step implemented is as follows:
The hunting zone of step 4-1: threshold value T is T ∈ [150,255], calculates successively with certain threshold value TiThe between class distance D determinedi。
Wherein, phFor image is assumed the pixel (p belonging to high intensity objecth>Ti), plFor image is assumed the pixel (p being not belonging to high intensity objectl<Ti), nhWhat represent is that brightness value is more than TiPixel number;nlWhat represent is that brightness value is less than TiPixel number.
Step 4-2: choose and there is maximum kind spacing DiTime corresponding TiFor threshold value.
Step 4-3: adopt this threshold value that gray level image carries out two classes (being divided into thermal objects, non-thermal objects).
Step 4-4: to being identified as thermal objects in previous step image, the method adopting connected region to extract, extract all connected regions and go forward side by side line flag.Namely the minimum enclosed rectangle that the ratio of each connected region is 2:1 thinks potential pedestrian region that may be present, is referred to as pedestrian and assumes.This step can exist some non-pedestrian region (such as there is the region of vehicle trees) is also comprised wherein, accordingly, it would be desirable to the pedestrian grader C by being obtained by step 3 is verified in the next step.
Step 5: treating in step 4 is judged, and pedestrian assumes with Haar characteristic vector HiCharacterize.Then being input to by this vector in the grader C that step 3 obtains, grader C exports and waits to judge the classification (pedestrian or non-pedestrian) that pedestrian assumes.By assuming to be verified to all pedestrians in step 4, namely can detect that all in width far infrared at a night image there is pedestrian.
Beneficial effects of the present invention:
The present invention is directed to the feature of far infrared road image at night, it is proposed that a kind of pedestrian detection method suitable in far infrared image at night.The method passes through pedestrian's hypotheses creation, and the SVM classifier based on Haar feature verifies that achieving the accurate of pedestrian in far infrared image at night detects.
Accompanying drawing explanation
Fig. 1 is the flow chart that the present invention proposes method;
Fig. 2 is the 10 Like-Fenton Oxidation schematic diagrams that the present invention chooses.
Detailed description of the invention
Below in conjunction with the drawings and specific embodiments, the invention will be further described.
As it is shown in figure 1, the flow chart proposing method for the present invention, comprise the steps:
Step 1: by manually choosing the picture of pedestrian in far infrared image at a large amount of night and non-pedestrian, be respectively labeled as positive negative sample.All positive negative samples are all normalized to length and width and is the size of 32x16 pixel.If positive and negative sample number is n altogether, positive number of samples is more than 5000, and negative sample number is more than 20000.
Step 2: by any one positive negative sample SiCharacterize by Haar feature, generate Haar characteristic vector Hi, (i=1,2 ..., n).Implement and include:
Step 2-1:Haar rectangular characteristic type is various, it is contemplated that pedestrian presents more horizontal and vertical lines feature in the picture, chooses 10 Like-Fenton Oxidation as shown in Figure 2.
Step 2-2: a certain Haar feature of some yardstick carries out corresponding calculating on sample image on some position can obtain a numerical value h.So, the Haar feature calculation numerical value of dissimilar, yardstick and position just constitutes a multidimensional Haar characteristic series vector H of certain samplei, (i=1,2 ..., n).
Step 3: this step is classifier training.The Haar characteristic vector H that will generate in step 2iIt is input in SVM classifier and is trained, can train and obtain vehicle classification device C.Concrete, by the characteristic vector H of the positive and negative sample image described in step 2iBeing trained with SVM, the kernel function of SVM adopts the most frequently used RBF-gaussian kernel function.
Step 4: read in far infrared image Img at actual night, and obtain pedestrian by adaptive threshold T and would be likely to occur region.In far infrared image under night-environment, part thermal objects is (such as the position such as pedestrian head, trunk, also vehicle is included) obvious high brightness is presented because temperature is higher, and in road, most object such as the road surface in image, number presents low-light level because temperature is low.Therefore, a suitable threshold value T can be passed through to be split.Threshold value determines that step is as follows:
The hunting zone of step 4-1: threshold value T is T ∈ [150,255], calculates successively with certain threshold value TiThe between class distance D determinedi。
Wherein, phFor image is assumed the pixel (p belonging to high intensity objecth>Ti), plFor image is assumed the pixel (p being not belonging to high intensity objectl<Ti), nhWhat represent is that brightness value is more than TiPixel number;nlWhat represent is that brightness value is less than TiPixel number.
Step 4-2: choose and there is maximum kind spacing DiTime corresponding TiFor threshold value.And adopt this threshold value that gray level image (i.e. aforementioned far infrared image Img) is carried out two classes (thermal objects, non-thermal objects).
Step 4-3: to being identified as thermal objects in previous step image, the method adopting connected region to extract, extract all connected regions and go forward side by side line flag.Namely the minimum enclosed rectangle of the 2:1 of each connected region thinks potential pedestrian region that may be present, is referred to as pedestrian and assumes.This step can exist some non-pedestrian region (such as there is the region of vehicle) is also comprised wherein, accordingly, it would be desirable to the pedestrian grader C by being obtained by step 3 is verified in the next step.
Step 5: will wait to judge that pedestrian assumes with Haar characteristic vector HiCharacterize.Then being input to by this vector in the grader C that step 3 obtains, grader C exports and waits to judge the classification (pedestrian or non-pedestrian) that pedestrian assumes.By assuming to be verified to the pedestrian in step 4, detect all in width far infrared at a night image there is pedestrian.
The above is only used for describing technical solution of the present invention, the protection domain being not intended to limit the present invention, it will be appreciated that under flesh and blood of the present invention and spirit premise, changed and retouching etc. all will in scope.
Claims (5)
1. the pedestrian detection method in a night far infrared image, it is characterised in that including: the step of the step of sample training and image detection;
The step of described sample training includes as follows:
Step 1: by manually choosing the picture of pedestrian in some width far infrared at night images and non-pedestrian, be respectively labeled as positive sample and negative sample, all positive samples and negative sample all normalized to length and width and is the size of 32x16 pixel;
Step 2: by any one positive sample or negative sample SiCharacterize by Haar feature, generate Haar characteristic vector Hi, i=1,2 ..., n, n is positive number of samples and negative sample number summation;
Step 3: utilize grader to be trained, particularly as follows: the Haar characteristic vector H that will generate in step 2iIt is input in SVM classifier and is trained, can train and obtain vehicle classification device C;
The step of described image detection includes as follows:
Step 4: read in far infrared image Img at actual night, and obtain pedestrian region that may be present by adaptive threshold T, is defined as pedestrian and assumes;
Step 5: treating in step 4 is judged, and pedestrian assumes with Haar characteristic vector HiCharacterize, then by this vector HiBeing input in the grader C that step 3 obtains, grader C exports and waits to judge that classification that pedestrian assumes is as pedestrian or non-pedestrian;
By assuming to be verified to all of pedestrian in step 4, namely can detect that the pedestrian of all existence in width far infrared at a night image.
2. the pedestrian detection method in far infrared image at a kind of night according to claim 1, it is characterised in that generate Haar characteristic vector H described in step 2iStep include:
Step 2-1: choose 10 Like-Fenton Oxidation;
Step 2-2: the Haar feature calculation numerical value of dissimilar, yardstick and position constitutes a multidimensional Haar characteristic series vector H of certain samplei, i=1,2 ..., n.
3. the pedestrian detection method in far infrared image at a kind of night according to claim 1, it is characterised in that the step that adaptive threshold T described in step 4 is determined includes:
The hunting zone of step 4-1: threshold value T is T ∈ [150,255], calculates successively with certain threshold value TiThe between class distance D determinedi:
Wherein, phFor image is assumed the pixel belonging to high intensity object, and ph>Ti, plFor image is assumed the pixel being not belonging to high intensity object, and pl<Ti, nhWhat represent is that brightness value is more than TiPixel number;nlWhat represent is that brightness value is less than TiPixel number;
Step 4-2: choose between class distance DiFor T corresponding time maximumiFor threshold value.
4. the pedestrian detection method in far infrared image at a kind of night according to claim 3, it is characterised in that the acquisition methods that pedestrian described in step 4 assumes includes:
Step 4-3: utilize threshold value T that gray level image carries out two classizationes and process, be divided into thermal objects and non-thermal objects;
Step 4-4: for being identified as thermal objects in step 4-3, adopt the method that connected region is extracted, extracting all connected regions to go forward side by side line flag, namely the minimum enclosed rectangle that the ratio of each connected region is 2:1 thinks potential pedestrian region that may be present, is pedestrian and assumes.
5. the pedestrian detection method in far infrared image at a kind of night according to claim 1, it is characterised in that described positive number of samples is more than 5000, and described negative sample number is more than 20000.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610119259.7A CN105787456A (en) | 2016-03-02 | 2016-03-02 | Method for detecting pedestrians in night far infrared images |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610119259.7A CN105787456A (en) | 2016-03-02 | 2016-03-02 | Method for detecting pedestrians in night far infrared images |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105787456A true CN105787456A (en) | 2016-07-20 |
Family
ID=56386932
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610119259.7A Pending CN105787456A (en) | 2016-03-02 | 2016-03-02 | Method for detecting pedestrians in night far infrared images |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105787456A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106919930A (en) * | 2017-03-10 | 2017-07-04 | 成都智锗科技有限公司 | A kind of low resolution infrared image parahypnosis situation determination methods |
CN108416254A (en) * | 2018-01-17 | 2018-08-17 | 上海鹰觉科技有限公司 | A kind of statistical system and method for stream of people's Activity recognition and demographics |
CN110866422A (en) * | 2018-08-28 | 2020-03-06 | 天津理工大学 | Two-dimensional false pedestrian recognition method based on combination of light field imaging and LBP (local binary pattern) and SVM (support vector machine) |
CN112422915A (en) * | 2020-11-18 | 2021-02-26 | 珠海格力电器股份有限公司 | Method and device for monitoring number of people, electronic equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101226597A (en) * | 2007-01-18 | 2008-07-23 | 中国科学院自动化研究所 | Method and system for recognizing nights pedestrian based on thermal infrared gait |
CN103198332A (en) * | 2012-12-14 | 2013-07-10 | 华南理工大学 | Real-time robust far infrared vehicle-mounted pedestrian detection method |
CN103279741A (en) * | 2013-05-20 | 2013-09-04 | 大连理工大学 | Pedestrian early warning system based on vehicle-mounted infrared image and working method thereof |
CN103902976A (en) * | 2014-03-31 | 2014-07-02 | 浙江大学 | Pedestrian detection method based on infrared image |
CN104091171A (en) * | 2014-07-04 | 2014-10-08 | 华南理工大学 | Vehicle-mounted far infrared pedestrian detection system and method based on local features |
-
2016
- 2016-03-02 CN CN201610119259.7A patent/CN105787456A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101226597A (en) * | 2007-01-18 | 2008-07-23 | 中国科学院自动化研究所 | Method and system for recognizing nights pedestrian based on thermal infrared gait |
CN103198332A (en) * | 2012-12-14 | 2013-07-10 | 华南理工大学 | Real-time robust far infrared vehicle-mounted pedestrian detection method |
CN103279741A (en) * | 2013-05-20 | 2013-09-04 | 大连理工大学 | Pedestrian early warning system based on vehicle-mounted infrared image and working method thereof |
CN103902976A (en) * | 2014-03-31 | 2014-07-02 | 浙江大学 | Pedestrian detection method based on infrared image |
CN104091171A (en) * | 2014-07-04 | 2014-10-08 | 华南理工大学 | Vehicle-mounted far infrared pedestrian detection system and method based on local features |
Non-Patent Citations (1)
Title |
---|
钟令: "《基于红外图像的车辆前方行人识别与跟踪技术研究》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106919930A (en) * | 2017-03-10 | 2017-07-04 | 成都智锗科技有限公司 | A kind of low resolution infrared image parahypnosis situation determination methods |
CN108416254A (en) * | 2018-01-17 | 2018-08-17 | 上海鹰觉科技有限公司 | A kind of statistical system and method for stream of people's Activity recognition and demographics |
CN110866422A (en) * | 2018-08-28 | 2020-03-06 | 天津理工大学 | Two-dimensional false pedestrian recognition method based on combination of light field imaging and LBP (local binary pattern) and SVM (support vector machine) |
CN112422915A (en) * | 2020-11-18 | 2021-02-26 | 珠海格力电器股份有限公司 | Method and device for monitoring number of people, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107891808B (en) | Driving reminding method and device and vehicle | |
Chen et al. | Turn signal detection during nighttime by CNN detector and perceptual hashing tracking | |
Kurihata et al. | Rainy weather recognition from in-vehicle camera images for driver assistance | |
Siogkas et al. | Traffic lights detection in adverse conditions using color, symmetry and spatiotemporal information | |
CN109190523B (en) | Vehicle detection tracking early warning method based on vision | |
CN103034843B (en) | Method for detecting vehicle at night based on monocular vision | |
CN107886034B (en) | Driving reminding method and device and vehicle | |
CN105574552A (en) | Vehicle ranging and collision early warning method based on monocular vision | |
Chen et al. | Nighttime brake-light detection by Nakagami imaging | |
Li et al. | Nighttime lane markings recognition based on Canny detection and Hough transform | |
CN104881661B (en) | Vehicle checking method based on structural similarity | |
CN105787456A (en) | Method for detecting pedestrians in night far infrared images | |
CN106679633A (en) | Vehicle-mounted distance measuring system and vehicle-mounted distance measuring method | |
CN106845458B (en) | Rapid traffic sign detection method based on nuclear overrun learning machine | |
CN106919939A (en) | A kind of traffic signboard Tracking Recognition method and system | |
CN104992160B (en) | A kind of heavy truck night front vehicles detection method | |
Bi et al. | A new method of target detection based on autonomous radar and camera data fusion | |
Meis et al. | Reinforcing the reliability of pedestrian detection in far-infrared sensing | |
Wen et al. | A rear-vehicle detection system for static images based on monocular vision | |
Chen | Road vehicle recognition algorithm in safety assistant driving based on artificial intelligence | |
CN104268859A (en) | Image preprocessing method for night lane line detection | |
CN108229447B (en) | High beam light detection method based on video stream | |
CN105303159A (en) | Far-infrared pedestrian detection method based on distinctive features | |
CN104966064A (en) | Pedestrian ahead distance measurement method based on visual sense | |
CN105718893A (en) | Car tail light pair detecting method for night environment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160720 |
|
RJ01 | Rejection of invention patent application after publication |