TW201044008A - Night time pedestrian detection technology - Google Patents

Abstract

The invention discloses a night time pedestrian detection system and method. The system of the invention includes an emitter, an image-capturing apparatus and a processing module. The emitter emits an infrared light toward the first direction, and the image-capturing apparatus receive the reflection of the infrared light from the first direction. Additionally, the processing module receives the infrared image from the image-capturing apparatus, and segments the infrared image into a plurality of image regions. The processing module determines a candidate region on the infrared image according to the contrast level between each of the image regions. Furthermore, the processing module determines if the candidate region includes a pedestrian by a cascaded AdaBoost classifier using an augmented histogram of orientation gradient of said candidate region.

Description

201044008 VI. Description of the Invention: [Technical Field of the Invention] The present invention relates to an image-based Langerine detection system and method, and in particular, to a vehicle-type imagery night pedestrian detection system and method. [Prior Art] When driving at night, the driver must rely on the illumination of the street lamp and the headlights of the vehicle to identify the road ahead to ensure safe driving. However, when external factors such as heavy rain or heavy fog, or money level and visual acuity, are present, the driver may easily overlook the pedestrians, obstacles, etc. in front of him and cause an accident. Therefore, methods and systems for night pedestrian detection through image analysis have been proposed. The nighttime pedestrian detection methods in the prior art can be mainly divided into two categories: Appearance-based and M〇ti〇n_based methods. The exterior night pedestrian detection method uses multiple pedestrian appearance images to train the Classifier, allowing the classifier to summarize several image features from a variety of pedestrian appearance images. Then, when the infrared image capturing device captures the infrared image, the trained classifier can be applied to determine whether the infrared image matches the summarized image feature by using a standard pattern recognition method (Pattem Recognition). If any, then judge as a possible pedestrian. Commonly used Histogram of Orientation Gradient (HOG), Image Wavelet 201044008 (Wavelet) 'Shape lntensity, etc.' and common classifiers have adaptive reinforcement (adaptive) Boosting) classifier AdaB〇()St, support vector machine_Port maehine, WM) classifier, neural network

Network, NN) classifiers, etc. In addition, the sporty night pedestrian side method is mainly to learn the image shape caused by the movement of pedestrians (4) (4), for example, the society is mixed (3) (10) Ca_) Ο

Hold, detect the object of the image, and make sure that the image obtained by the camera is in the same place, and then establish the similarity with the image of the next-time riding image, and finally determine which object is an independently moving object. However, the far-infrared camera used in the current night pedestrian detection system is more complicated, more bulky and consumes electricity, and is more difficult to be widely used. Correction, the existing night pedestrians _ New Zealand is more like the eye (9) sound version (10) as the design, according to 'therefore, it faces the problem of low price, installation _, and complicated calculation. SUMMARY OF THE INVENTION This aspect of the present invention provides an image-type nighttime field 涮糸, 4 to solve the problems in the prior art. Including, Γ Γ + ΐ 实 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , The emitter emits infrared light from the outside and the outside, and the image capturing device can capture the infrared image of the W outer line in the material-direction. The outer knot is connected to the home image capturing device, and the red infrared image is subdivided into a plurality of image blocks, and the infrared image is determined according to the contrast of each image block and the adjacent image block to 5 201044008. Selecting the region 'and according to the enhancement direction gradient of each candidate region is straightforward whether the at least one candidate block contains a pedestrian image. Ice, another – Fan • provides a method of night-time pedestrian tax measurement to solve problems in the prior art.

According to the specific implementation method of the present invention, the image-type night-breaking detection method comprises: (a) emitting an infrared ray in a direction toward the first direction; (b) absorbing an infrared ray image formed by the reflection in the direction of the distance - (8) Connecting the external image, dividing the infrared image into a plurality of image blocks, determining at least a candidate region of the infrared image according to the contrast of each image unit and the adjacent image block; (d) generating one of each candidate region The enhanced directional gradient straight feature; and (8) the continuation segment AdaB_ classifier confirms whether the y candidate block includes a ^^^ pedestrian image. The advantages and spirit of the present invention will be further understood from the following description of the invention. [Embodiment] The present invention provides an image nighttime pedestrian detection system and method, which are described below in accordance with several embodiments of the present invention. x Please refer to FIG. 1 , FIG. 2 , FIG. 3A and FIG. 3B together to illustrate an image night pedestrian detection system according to an embodiment of the present invention; FIG. 2 is a schematic diagram of the vehicle; FIG. A flowchart of an image-based night pedestrian detection method according to an embodiment of the present invention; FIG. 3A and FIG. 3B are schematic diagrams showing an infrared image according to an embodiment of the present invention. The night vision detection system 1 of the present invention includes a transmitter ίο, an image capture voucher 12, a processing module 14 and a display 16. 6 201044008 The step-initiator 10 emits infrared rays (10) toward the first direction 〇1 (step S50). In the present embodiment, the infrared ray is near infrared ray. The image capturing device 12 picks up the infrared ray 1 and reflects the reflected infrared ray image 2 in the first direction m (step S52). As shown in Fig. 1, the transmitter 1 and the image capturing device 12 can be disposed at the front end of the automobile 9. In addition, the processing module 14 is connected to the image capturing device 12 for receiving the infrared ray 2, and the heterogeneous image is 2 years old into a plurality of image blocks (for example, but not limited to pixels), according to each The contrast between the image block and the adjacent image block is determined as at least the candidate block 2 红外线 on the infrared image 2 (step S54). And the processing module 14 can be based on one of the candidate blocks 2 enhanced direction gradient histogram (Augmented Histograms of Orientation)

The Gradient, AHOG) feature' and by the staged AdaB〇〇st classifier confirms whether at least one candidate block 20 contains a pedestrian image 200 (step S54). Further, the display 16 is coupled to the processing module 14 for displaying the infrared image 2 and the location of the candidate block 20 containing the pedestrian image 2〇〇. In practice, the display 16 can be placed in front of or in the peripheral area of the driver's seat. Referring to FIG. 3A, FIG. 3B, FIG. 4 and FIG. 5, FIG. 4 is a functional block diagram of a processing module according to an embodiment of the present invention; FIG. 5 is not a step S54 in FIG. Detailed flow chart. In the present embodiment, the foregoing processing module 14 further includes a candidate block generating unit 14 〇 'selecting unit 142, a sorting unit 144, and a verifying unit 146. In practice, the candidate block generation early element 140, the selection unit 142, the classification unit 144, and the verification unit 146 may be separately disposed on the circuit board, or the units may be integrated into an integrated circuit chip. The candidate block generating unit 140 divides the infrared image 2 into a plurality of image early positions and uses the adjacent image information to obtain the possible foreground position and size in a multi-adaptive threshold segmentation manner. The candidate block 2 that may include the pedestrian image (step S540) is in practical application. 'We assume that the night pedestrian will appear near the high contrast area in the near infrared. The candidate block generating unit 140 transmits a threshold value of a plurality of brightnesses to determine whether each image unit belongs to the foreground or the background. My reference

Dong's paper (see ’ Jianfei Dong, Junfeng Ge and Yupin)

Luo, "Nighttime Pedestrian Detection with Near Infrared using Cascaded Classifiers, IEEE International Conference 〇n Image iV (10) tearing PP 185-188, 2007) 'and modify the range definition of the neighborhood of the method, from the original single scan line Multiple scan lines, defined as [Formula 1], and further use Dong's [Formula 2] to define two thresholds, ❹ and Thigh. i+*/2 y+Jfc/2 Σ Σ rr (zi\ _ x^-kny^jk/l_ [Formula 1] [Formula 2] where A is the size of the image block in pixels; 0 is a constant Moreover, the judgment rules of the scene and the scene are as defined in the Dong paper [Formula 3] 8 201044008 P(U)^B, turn, ί) <Τ』, β

311(1 P(i~l,j) eF or Pd-l,j) eB

[Formula 3] The other documents are in the full text, where the 'corpse represents the foreground; and 5 represents the background. After 〇 4, the various image units that belong to the foreground are combined to form a white dotted frame in the 2Q _ 4 B of the pedestrian. In addition, the selecting unit 142 is connected to the candidate block generating unit 14A for performing step-by-step screening according to the block characteristics of each candidate block 2G (step S542), because the pedestrian usually has a fixed position, and the size is some In a range, the selecting unit 142 may filter out the candidate block 20 that is too large or too small, the position is in the sky, and the aspect ratio is incorrect according to the size, position, length and width ratio, etc. of the candidate block 20. The next possible candidate block is 2〇. In the inter-application, based on the speed of the algorithm, the order of the candidate blocks can be set as follows: (1) Filter out blocks that are too large or too small, for example, one tenth larger than the image. The area with less than one thousandth of the image is not considered: Since this type of block contains almost no target (ie, pedestrian) to be detected, and only one integer comparison is needed to calculate the result, Put the judgment in the first-order position; (2) filter out the area where the position is unreasonable, for example, leave only the area below the vanishing line: this operation also requires only one integer size comparison, but usually appears in an unreasonable position. There are fewer candidate blocks, so the first unreasonable size of the 9 201044008 block can reduce the number of candidate blocks to be compared in this step, thereby improving the speed of the difference, and (3) filtering out the length and width are unreasonable. Blocks, for example, only leave the aspect ratio of two to one: this step requires a floating point division and a floating point size comparison, so it takes the most g between system executions, so It is placed in the last step carried out. Of course, in practice, the selecting unit 142 of the present invention can also perform _ selection through other block characteristics, and the filtering order can be adjusted as appropriate, and is not limited to the order exemplified herein. ,

Referring to FIG. 6, FIG. 6 is a schematic diagram showing the infrared image in FIG. 3B processed through the foregoing steps. As shown in the figure, through the foregoing steps, only two candidate blocks remain on the infrared image 2, and other candidate blocks have been excluded. Further, the feature unit 144 is connected to the selection unit 142 for filtering according to the enhanced direction gradient histogram of the remaining candidate blocks (step ^$544). Referring to FIG. 7, FIG. 7 illustrates one specific according to the present invention. The construction of the expansion histogram of the embodiment is a schematic diagram. The calculation process of the 'symmetry weight ^ window is as follows: suppose the side view t has a vertical axis of symmetry between the towels, the value of the symmetry tree # two corresponding area towels, the relative degree of relative pixels 'in order to reduce The image + target rotation and the impact of the touch adopt the single-pixel similarity, and choose to consider the average of the similarity of the pixel.凊 See Figure VIII. Figure 8 shows the schematic diagram of the construction process of the similarity. Further, the value of the similarity can be obtained by the following [Formula 4]. 201044008 [Formula 4] similar {x, y) = ^f\x, y) 'xm{x, y) where / represents the block after the original block is flipped, and w represents the block after the original block is mirrored . Therefore, the symmetry weight value can be derived from the following [Formula 5]. I] similar {x \yr) [Formula 5]

SymWeight {x,y) = (less (four) (eight less) -

Nk(x>y) = {(x,,y')eZ^ \x-^<x'<x + ^ andy-^<y'<y + ^}

Where iVjtO, less) represents (x, y) the pixels of all neighboring regions, A: indicates the size of this region and #(A^(x, _y)) is the sum of all the pixels in this region. The second enhanced part is the density of the gradient strength. For a certain characteristic block, the value can be derived from the following [Formula 6]. Densityblock = #({(x,^) e block I mag{x,y) > threshold]^ #({(x,^) € block I mag(x,y) ^ [Formula 6]

The last enhanced part is the contour of the human figure. For a feature block, its value can be derived from [Formula 7] and [Formula 8] below. [Formula η Σ mag(x, y)xEdist(x, y)

Dist = -

1 L

Edist{x,y) = yj(xx)2 +(yy)2 [Equation 8] where five represents the distance between (X,;;) and its humanoid center & force in Euclidean space, this distance will Multiplying the gradient strength of this pixel 11 201044008 'The resulting value is normalized to between 〇 and 1. The construction process of the enhanced direction gradient histogram has been exposed in the literature (see 'Zhuang Zhenxun's "Using Monocular Vision Enhanced Directional Gradient Histogram for Multi-Person Detection", July 30, 2008) This document is incorporated herein by reference in its entirety. In the affiliation, we use a classification unit 146 that combines Machine Learning and Knowledge-based. Using a large number of pedestrian images and non-pedestrian images, a strong classifier is learned through the Adaboost method. In particular, the enhanced directional gradient histogram features with discriminative power are picked out during the learning process and combined to form a strong classifier. The enhanced directional gradient histogram feature is one of the features developed in the original directional gradient histogram that enhances human symmetry, contour distance, and density of gradient strength, and is particularly suitable for characterizing pedestrians. The candidate block feature information calculated by the feature unit 144 is transmitted to the classifying unit 146. By using the AdaBoost strong classifier and candidate block feature information of the machine learning method, it can be determined whether the feature information belongs to a pedestrian, and therefore the classifying unit 146 classifies whether the candidate block has a pedestrian, and if so, confirms that the candidate block includes Pedestrian image (step S546). Referring to FIG. 9 , FIG. 9 is a schematic diagram of the infrared image in FIG. 6 processed through the foregoing steps. As shown, through the foregoing steps, only one candidate block remains on the infrared image 2, and the candidate block contains a pedestrian image. 12 201044008 = In the meantime, the image type night pedestrian _ system of the present invention can be fixedly installed, such as a building, a six-two or other suitable location, in addition to the vehicle; In addition, the image night frame system of the present invention can be used for identification of animal tools in addition to night pedestrian identification, and the image of the utility model is compared with the prior art. For a single near-infrared camera, system requirements = Ο This is a low-valley acquisition and erection. _ - Having succeeded by the detailed description of the preferred embodiment above, ===_: conversely, =. Therefore, the application of the present invention; the interpretation of the most comfortable: =: covers all of the available = 〇 13 201044008 [Simplified illustration of the drawings] FIG. 1 illustrates an image night pedestrian detection according to an embodiment of the present invention. A schematic diagram of the system set up on the car. 2 is a flow chart showing an image night pedestrian detection method according to an embodiment of the present invention. FIG. 3A and FIG. 3B show a schematic diagram of an infrared image according to an embodiment of the present invention. 4 is a functional block diagram of a processing module in accordance with an embodiment of the present invention. FIG. 5 is a detailed flow chart of step S54 in FIG. Figure 6 is a schematic diagram showing the infrared image in Figure 3B processed through the steps of Figure 5. Figure 7 is a flow chart showing the construction process of the enhanced directional gradient 〇 histogram according to an embodiment of the present invention. FIG. 8 is a schematic diagram showing the construction process of the aforementioned similarity. Figure 9 is a schematic diagram showing the infrared image in Figure 6 after being processed through the steps of Figure 7. [Main component symbol description] 10: Transmitter 12: Image capturing device wo: Candidate block generating unit 1: Image type night pedestrian detecting system 100: Infrared Η Processing module 14 144: Feature unit 16: Display 20: Candidate Block 9: Car 201044008 142: Selection Unit 146: Classification Unit 2: Infrared Image 200: Pedestrian Image S50~S54, S540~S546 ··Process Step

15

Claims (1)

201044008 VII. Patent application scope: i. An image-type night pedestrian detection system, comprising: a shooting person used to transmit a pair of infrared rays to a younger one; an image capturing device for capturing the infrared light in the first An infrared image formed by reflection in one direction; ―, the module is connected to the image capturing device for receiving the infrared image, dividing the infrared image into a plurality of image blocks, and determining the infrared image according to the contrast between each image block and the adjacent image block. And at least one candidate block, and according to one of each candidate block, an augjj^ted Histograms of OnentationGradient (AH0G) feature is confirmed, and the at least one candidate block is confirmed by the staged AdaBoost classifier. Contains a pedestrian image. The image-based model of the pedestrian-side system of the first item, wherein the processing module further comprises: a candidate block generating unit that divides the infrared image into a plurality of image units and cuts with multiple adaptive valves ( a multi_adaptive threshold segmentation manner, generating at least one first candidate block; a selecting unit, connecting the candidate block generating unit, and selecting, by the rib, at least one second candidate block according to at least one block characteristic of each of the first candidate blocks a single 70' is connected to the selection unit, and the enhanced direction gradient histogram feature information is calculated according to each of the second candidate blocks to output at least the enhanced direction gradient histogram feature information of the second candidate block; and St, scale The levy unit is classified by the stage peach (10) =: the fourth quarantine block, the at least one enhanced direction gradient histogram characteristic is not a pedestrian, and if so, the 16th 201044008 candidate block is included Pedestrian imagery. 3. The image type night pedestrian detection system of claim 2, wherein the at least one block characteristic comprises a size, an aspect ratio of the block, and a position on the infrared image. 4. The image type night pedestrian detection system according to claim 2, wherein the classification unit is a Strong Classifier. 5. The image type night pedestrian detection system according to claim 2, wherein the candidate block generating unit, the selecting unit and the sorting unit are integrated in an integrated circuit chip. 6. For example, the image-type night pedestrian detection system described in claim 1 is characterized in that the enhanced direction gradient histogram is based on a one-direction gradient histogram and enhances the geometrical characteristics of the human body. 7. The image type night pedestrian detection system according to claim 6, wherein the human geometric characteristics include symmetry, contour distance, and gradient strength. Mountain 0. The image-type nighttime pedestrian copper system of claim 1 of the patent application further includes: a display for displaying the infrared image and the position of the candidate block including the pedestrian image. y, the image-type night pedestrian survey system described in the first paragraph of the patent range, the infrared system-near infrared ray in the basin.八10、-Image type night pedestrian detection method, comprising the following steps: (a) emitting _IR in the direction of the first direction; () absorbing infrared rays in the first direction and forming one of the infrared rays 17 201044008 line (C) receiving the infrared image, dividing the infrared image into a plurality of image blocks 'determining at least one candidate block on the infrared image according to the contrast of each image block and the adjacent image block; and (4) according to An Augmented Histograms of Orientation Gradient (AHOG) feature of each of the candidate blocks confirms whether the at least one candidate block contains a pedestrian image by the staged AdaBoost classifier. 11. The image type night pedestrian insult method according to the first aspect of the patent application, wherein the step (d) further comprises the following steps: (dl) dividing the infrared image into a plurality of image units, and multiplexing a multi-adaptive threshold segmentation method, generating at least one first candidate block; (d2) screening at least one second candidate block according to at least one block characteristic of each first candidate block; (d3) according to each The first candidate block calculates the feature information of the enhanced direction gradient histogram feature information outputting a second candidate block; and (d4) classifying the second candidate block by the staged AdaBoost strong classifier Whether it belongs to a pedestrian' If yes, it is believed that the second candidate block contains the pedestrian image. '!2, ^ Please refer to the image type Langren detection method described in the item n of the patent, wherein the at least one block characteristic in the basin includes the size of the block and the position on the infrared image. And eight (4) 13. The image-type night pedestrian prescribing method described in item 1 of the profit range, whose = strong direction gradient histogram is based on a one-direction gradient histogram and enhances the geometric characteristics of the human body. A. The image-type night pedestrian detection method described in claim 13 of the patent application, 18 201044008 】 the geometrical characteristics of the human body include the difficulty of the contour, the contour distance minus the intensity of the gradient, and the image night described in the 1G item. In the method of breaking the human side, the progress includes the following steps: Infrared image and the selection block containing the image of the Saki people 16, = the image-type night described in the first application of the patent range. '仃人 detection method, its 〇 19