Traffic Light Recognition Method, System and Storage Medium Based on Binocular Camera

Abstract

The present invention relates to the field of vehicle intelligent driving technology, and specifically relates to a traffic light recognition method, system and storage medium. It mainly obtains a long-focus target image and a short-focus target image of a traffic light and performs image fusion; obtains a fusion target box and fusion color information of the fusion image; obtains regions of interest of the long-focus target image and the short-focus target image and respectively obtains long-focus / short-focus weights, and finally determines an output color and an output target box to obtain a traffic light recognition result. This method uses a traffic light target recognition method combining long focus and short focus, integrating the advantages of better detection effect of long focus on distant targets and wider detection range of short focus, and overcoming the disadvantage of poor detection ability of a monocular camera. It greatly improves the intelligence level of the autonomous driving system and has better application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle intelligent driving, and particularly to a traffic light recognition method, system and storage medium. Background Art

[0002] With the continuous development of social economy and science and technology, automobiles, as a means of transportation, have become an integral part of people's lives. Safe driving and intelligent driving have become the directions pursued by people. Traffic light recognition in autonomous driving is an important part of the road detection module in autonomous driving. It is mainly used to identify vehicle traffic lights and pedestrian traffic lights on the road, for vehicle driving control. Detecting different traffic lights will give different feedback reminders to the driver.

[0003] In the existing traffic light target detection methods, it is mainly achieved by collecting road images through a camera and performing recognition and analysis. For example, the Chinese invention patent document with the publication number CN 111582216 A discloses an unmanned vehicle traffic light recognition system and method, which proposes to solve the problem of low accuracy of traffic light recognition due to image blurring based on camera detection and color space transformation, and realizes the recognition of the position of traffic lights in unmanned driving.

[0004] Furthermore, most traffic light target detection methods use a monocular camera in combination with a specific algorithm. However, this method of software and hardware cooperation often has problems such as a fixed focal length, a relatively blurred captured image and inaccurate color recognition. As a result, in the subsequent process of recognizing traffic light targets, it will increase the probability of misjudgment and missed recognition, and reduce the safety and intelligence of the autonomous driving system. Summary of the Invention

[0005] The purpose of the present invention is to provide a traffic light recognition method, system and storage medium based on a binocular camera. It uses a traffic light target recognition method that combines a long focal length and a short focal length, synthesizes the advantages of better detection effect of the long focal length on distant targets and a wider detection range of the short focal length, and overcomes the disadvantage of poor detection ability of the monocular camera. It greatly improves the intelligence level of the autonomous driving system and has better application prospects.

[0006] In order to achieve the above purpose, the technical solutions adopted by the present invention are specifically as follows:

[0007] A traffic light recognition method based on a binocular camera, characterized in that the method is specifically as follows:

[0008] S01. Set up a binocular camera, where the binocular camera consists of a long - focal - length camera and a short - focal - length camera respectively, which are used to obtain a long - focal - length target image A1 and a short - focal - length target image B1 of the traffic light; extract the original color information S and the original target box K of the long - focal - length target image A1 and the short - focal - length target image B1 respectively;

[0009] S02. Perform image fusion on the long - focal - length target image A1 and the short - focal - length target image B1 to generate a fused image C; obtain the fused target box C K and the fused color information C S ;

[0010] S03. Obtain the long - focal - length region of interest ROI1 of the long - focal - length target image A1; obtain the short - focal - length region of interest ROI2 of the short - focal - length target image B1; respectively obtain the long - focal - length / short - focal - length weights according to the coincidence degree of their respective original target boxes K and the regions of interest;

[0011] S04. Calculate the coincidence degree between the fused target box C K and the original target boxes K of the long - focal - length / short - focal - length, and determine the output color and the output target box according to the original color information S of the long - focal - length / short - focal - length, the fused color information C of the traffic light S and the long - focal - length / short - focal - length weights;

[0012] S05. Package the output color and the output target box into an identification package and send it to the vehicle navigation system to complete the traffic - light recognition.

[0013] Thus, install a long - focal - length camera and a short - focal - length camera on the test vehicle to form a binocular camera system, and complete the internal and external parameter calibration using a special bracket during installation. The binocular camera system is controlled by a controller with a GPU to perform image shooting and data acquisition, and the controller pre - processes the collected data and performs neural network inference. The vehicle navigation system needs to receive the output color and the output target box of the traffic - light data to assist the vehicle in unmanned driving.

[0014] The long - focal - length camera is used to obtain the long - focal - length target image A1, and the short - focal - length camera is used to obtain the short - focal - length target image B1. The controller respectively performs traffic - light color recognition and original target box acquisition on the long - focal - length target image A1 and the short - focal - length target image B1. Among them, the original target box K in the long - focal - length target image A1 is used to locate the position of the traffic light in the long - focal - length target image A1; the original target box K in the short - focal - length target image B1 is used to locate the position of the traffic light in the short - focal - length target image B1. The controller automatically determines the position of the original target box K using the extraction algorithm stored in its internal memory. This extraction algorithm can be a feature - position recognition algorithm in the prior art, etc. The respective original color information S of the long - focal - length target image A1 and the short - focal - length target image B1 represents the color recognition results of the controller for them respectively, and this original color information S will be used as a reference object in subsequent judgments.

[0015] The processor extracts the coordinate pixel - block information of the long - focal - length target image A1 and the short - focal - length target image B1, and inputs the two into the fusion model network according to the mapping relationship for image fusion to generate the fused image C. The fusion model network can select, for example, an image - fusion technology based on a convolutional neural network (CNN). Or it can adopt the form of separately extracting the feature points of the long - focal - length / short - focal - length images and performing feature matching to form a homography matrix for fusion. In image fusion, edge detection, corner detection, and texture analysis and other methods can be used to extract the target traffic - light features in the image. During fusion, object detection is used to find the traffic lights jointly included in the long - focal - length target image and the short - focal - length target image, and deep - learning algorithms are used to classify and identify the features. Finally, the target traffic - light objects and feature information in the long - focal - length target image and the short - focal - length target image are fused together to generate a new fused image C.

[0016] Then, the fused target box C of the fused image C is obtained. K And the fused color information C S 。The method for selecting the fused target box C K can be the confidence - degree judgment method in the prior art. Then, the long - focal - length region of interest ROI1 of the long - focal - length target image A1 is obtained; the short - focal - length region of interest ROI2 of the short - focal - length target image B1 is obtained; the region of interest is a specific - size region at the center of the long - focal - length / short - focal - length image, which can be set to 1 / 4 or 1 / 5, etc. according to the area. The long - focal - length / short - focal - length weights are respectively obtained according to the overlapping degree between the original target box K and the region of interest. Specifically, if the overlapping degree between the original target box K of the long - focal - length target image A1 and the defined long - focal - length region of interest ROI1 is higher, the long - focal - length weight is larger, and the short - focal - length judgment method is the same.

[0017] After that, the overlapping degree between the fused target box C K and the original target box K of the long - focal - length / short - focal - length is calculated, and according to the original color information S of the long - focal - length / short - focal - length, the traffic - light fused color information C Sand the telephoto / short - focal - length weights to determine the output color and the output target box. Specifically, for example, if the coincidence degree between the fused target box C K and the original target box K of the telephoto image is greater, and the original color information S of the telephoto image is the same as the fused color information C of the traffic light S then the output color is based on the original color information S of the telephoto image; moreover, the position information of the output target box is obtained by weighting the original target box K of the telephoto image and the fused target box C K and the weight for weighting is based on the telephoto weight obtained in the above steps.

[0018] Finally, the output color and the output target box are packaged into a device result and sent to the vehicle navigation system to complete the traffic - light recognition.

[0019] In summary, this system uses a traffic - light target recognition method that combines telephoto and short - focal - length. It combines the advantages of better detection effect of telephoto for distant targets and wider detection range of short - focal - length, and overcomes the disadvantage of poor detection ability of a single - camera. It can reduce the probability of incorrect judgment of the traffic - light target and the traffic - light color by the vehicle - mounted navigation system, greatly improve the intelligence level of the autonomous - driving system, and has better application prospects.

[0020] As a preference of the present invention, in step S02, before fusing the telephoto target image A1 and the short - focal - length target image B1, it further includes an image pre - processing step: obtaining a telephoto - image pixel block according to the telephoto target image A1; obtaining a short - focal - length - image pixel block according to the short - focal - length target image B1.

[0021] Image pre - processing is a prerequisite for image fusion, which can make the images to be fused have, for example, the same size and contrast, and also includes steps such as image denoising to reduce noise interference in the image. Among them, obtaining the telephoto - image pixel block and the short - focal - length - image pixel block are preparations for pixel - level fusion. Pixel - level fusion means performing weighted summation on the pixels of the telephoto target image and the short - focal - length target image to generate a fused image C.

[0022] As a preference of the present invention, in step S03, the telephoto region of interest ROI1 occupies 1 / 4 of the total pixel area of the telephoto target image A1; the short - focal - length region of interest ROI2 occupies 1 / 5 of the total pixel area of the short - focal - length target image B1.

[0023] Taking the center point of the long - focal - length target image A1 or the short - focal - length target image B1, the long - focal - length region of interest ROI1 and the short - focal - length region of interest ROI2 are proportionally divided into rectangles. For example, the width of the long - focal - length target image A1 is W, the height is H, O is the center point of the long - focal - length target image A1, and the rectangular ROI1 region is a rectangular frame region centered on point O with a width of W / 2 and a height of H / 2. At this time, the long - focal - length region of interest ROI1 occupies 1 / 4 of the total pixel area of the long - focal - length target image A1; the short - focal - length region of interest ROI2 is the same.

[0024] In image processing, the region of interest is an area outlined in the processed image in the form of a square, circle, ellipse, irregular polygon, etc., which is the focus of image analysis. This area is circled for further processing. Using the region of interest to circle the target can reduce processing time and increase accuracy.

[0025] As a preference of the present invention, in step S03, when calculating the coincidence degree between the original target box K of the long - focal - length / short - focal - length and the corresponding region of interest to obtain the long - focal - length / short - focal - length weights respectively, a normalization operation is also included.

[0026] Thus, through data acquisition and automatic operation of the in - vehicle computer processor, the coincidence degree between the original target box K of the long - focal - length / short - focal - length and the corresponding region of interest can be obtained. Quantifying this coincidence degree can obtain the long - focal - length / short - focal - length weights, and performing a normalization operation on them to make the range of the weight factors belong to [0, 1] for the convenience of subsequent program operations.

[0027] As a preference of the present invention, in step S04, when determining the output color according to the original color information S of the long - focal - length / short - focal - length and the fused color information C of the traffic lights S a confidence - level determination operation is also included: when the original color information S of the long - focal - length / short - focal - length does not match the fused color information C of the traffic lights S the final traffic - light color is set to the original color information S of the party with a higher confidence level in the long - focal - length / short - focal - length image.

[0028] Both the long - focal - length target image A1 and the short - focal - length target image B1 have an independent confidence level. The confidence level refers to the probability that the error of the measurement result is within a certain range, which is called the confidence probability of the result, also known as the confidence level, and its value is between 0 and 1.

[0029] In S04, if the coincidence degree between the original target box K of the long - focal - length target image A1 and the fused target box C of the fused image C K is higher, and the original color information S of the long - focal - length target image A1 matches the fused color information C of the fused image C SIf they are the same, the final output color is based on the original color information S of the long - focal - length target image A1, and the position of the output target box is obtained by weighting the original target box K of the long - focal - length target image A1 and the fused target box C of the fused image C K according to the long - focal - length weight; If the original target box K of the long - focal - length target image A1 and the fused target box C of the fused image C K have a higher degree of overlap, but the original color information S of the long - focal - length target image A1 and the fused color information C of the fused image C S are different, then the final output color is based on the detection with a higher confidence in the long - focal - length target image A1 and the short - focal - length target image B1, and the position of the output target box is obtained by weighting the original target box K of the long - focal - length target image A1 and the original target box K of the short - focal - length target image B1 according to the long - focal - length weight.

[0030] A traffic - light recognition system based on a binocular camera, characterized in that the system comprises:

[0031] An image acquisition module, configured to acquire a long - focal - length target image A1 and a short - focal - length target image B1 of a traffic light; and extract the original color information S and the original target box K of the long - focal - length target image A1 and the short - focal - length target image B1 respectively;

[0032] An image fusion module, configured to perform image fusion on the long - focal - length target image A1 and the short - focal - length target image B1 to generate a fused image C; acquire the fused target box C K and the fused color information C S ;

[0033] A weight calculation module, configured to acquire the long - focal - length region of interest ROI1 of the long - focal - length target image A1; acquire the short - focal - length region of interest ROI2 of the short - focal - length target image B1; and respectively obtain the long - focal - length / short - focal - length weights according to the coincidence degree between the respective original target box K and the region of interest;

[0034] A data processing module, configured to calculate the coincidence degree between the fused target box C K and the original target box K of the long - focal - length / short - focal - length, and determine the output color and the output target box according to the original color information S of the long - focal - length / short - focal - length, the fused color information C of the traffic light S and the long - focal - length / short - focal - length weights;

[0035] A result sending module, configured to package the output color and the output target box into a recognition result and send it to the vehicle navigation system to complete traffic - light recognition.

[0036] Preferably, the image fusion module further includes an image preprocessing module, which is used to obtain a long - focal - length image pixel block according to the long - focal - length target image A1, and obtain a short - focal - length image pixel block according to the short - focal - length target image B1.

[0037] Preferably, the data processing module further includes a confidence determination module, which is used to set the final traffic - light color to the original color information S of the party with a higher confidence in the long - focal - length / short - focal - length image when the original color information S of the long - focal - length / short - focal - length and the fused color information C of the traffic light S do not match.

[0038] A readable storage medium applicable to an in - vehicle computer, on which there is a readable storage medium for executing any one of the above - mentioned traffic - light recognition methods based on a binocular camera.

[0039] In summary, the present invention has the following beneficial effects:

[0040] 1. The system uses a method of traffic - light target recognition that combines long - focal - length and short - focal - length, combines the advantages of better detection effect of long - focal - length on distant targets and wider detection range of short - focal - length, and overcomes the disadvantage of poor detection ability of a monocular camera. It can reduce the probability of misjudgment of traffic - light targets and traffic - light colors by the in - vehicle navigation system, greatly improve the intelligence level of the autonomous driving system, and has better application prospects.

[0041] 2. By dividing the region of interest, that is, defining the long - focal - length region of interest ROI1 and the short - focal - length region of interest ROI2, this region is convenient for further processing. Using the region of interest to define the target can reduce the processing time and increase the accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of this traffic - light recognition method;

[0043] Figure 2 is a structural block diagram of this traffic - light recognition system;

[0044] Figure 3 is a schematic diagram of the division of the region of interest. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] Hereinafter, the preferred embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Any person can implement the present disclosure in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to be able to fully convey the scope of the present disclosure to those skilled in the art.

[0046] As used herein, the term "comprising" and its variations mean open-ended inclusion, i.e., "including but not limited to". Unless otherwise specified, the term "or" means "and / or". The term "based on" means "at least partially based on". The term "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment".

[0047] In this embodiment, a binocular camera system composed of a long-focus camera and a short-focus camera (the focal length of the long-focus camera is 12 mm, and the focal length of the short-focus camera is 6 mm) is installed on the driverless minibus. During installation, the positions of the two cameras have been fixed using a special bracket and both internal and external parameter calibrations have been completed. At the same time, the vehicle navigation system includes a controller and a processing program that can run in the controller.

[0048] As Figure 1 shown, the long-focus camera is used to obtain a long-focus target image A1, and the short-focus camera is used to obtain a short-focus target image B1. Then, an image preprocessing step is performed. After synchronizing the image sizes, denoising, and adjusting the contrast of the images, the controller extracts the original color information S and the original target box K of the long-focus target image A1 and the short-focus target image B1 respectively. Specifically, the controller automatically determines the position of the original target box K using the extraction algorithm stored in its internal memory. This extraction algorithm can be a feature position recognition algorithm in the prior art, etc. The original color information S of the long-focus target image A1 and the short-focus target image B1 respectively represents the color recognition result of the controller for each of them, and this original color information S will be used as a reference object in subsequent judgments.

[0049] In addition, during the preprocessing process, a long-focus image pixel block will also be obtained according to the long-focus target image A1; a short-focus image pixel block will be obtained according to the short-focus target image B1. Then, the pixels of the long-focus target image and the short-focus target image are weighted and summed to perform pixel-level fusion to generate a fused image C. The processor extracts the coordinate pixel block information of the long-focus target image A1 and the short-focus target image B1, and inputs the two into the fusion model network according to the mapping relationship for image fusion to generate a fused image C. The fusion model network can select, for example, an image fusion technology based on a convolutional neural network (CNN). Or adopt the form of separately extracting the feature points of the long-focus / short-focus images and performing feature matching to form a homography matrix for fusion. During fusion, the traffic lights jointly included in the long-focus target image and the short-focus target image are found through object detection, and the features are classified and recognized based on a deep learning algorithm. Finally, the target traffic light objects and feature information in the long-focus target image and the short-focus target image are fused together to generate a new fused image C.

[0050] Then, the fused target box C of the fused image C is obtained K and the fused color information CS The selection method of the fused target box C K can be the confidence judgment method in the prior art. Then, obtain the long - focal - length region of interest ROI1 of the long - focal - length target image A1; obtain the short - focal - length region of interest ROI2 of the short - focal - length target image B1; the region of interest is a specific - size region at the center of the long - focal - length / short - focal - length image. In this embodiment, the long - focal - length region of interest ROI1 occupies 1 / 4 of the total pixel area of the long - focal - length target image A1; the short - focal - length region of interest ROI2 occupies 1 / 5 of the total pixel area of the short - focal - length target image B1. As Figure 3 shown, taking the center point of the long - focal - length target image A1 or the short - focal - length target image B1, the long - focal - length region of interest ROI1 and the short - focal - length region of interest ROI2 are proportionally divided into rectangles. For example, if the width of the long - focal - length target image A1 is W and the height is H, and O is the center point of the long - focal - length target image A1, the rectangular ROI1 region is a rectangular box region centered at point O with a width of W / 2 and a height of H / 2; at this time, the long - focal - length region of interest ROI1 occupies 1 / 4 of the total pixel area of the long - focal - length target image A1; the short - focal - length region of interest ROI2 is the same.

[0051] After that, obtain the long - focal - length / short - focal - length weights respectively according to the coincidence degree between their respective original target boxes K and the regions of interest; specifically, if the original target box K of the long - focal - length target image A1 has a higher coincidence degree with the defined long - focal - length region of interest ROI1, the long - focal - length weight is larger, and the short - focal - length judgment method is the same.

[0052] Then, calculate the coincidence degree between the fused target box C K and the original target boxes K of the long - focal - length / short - focal - length, and determine the output color and the output target box according to the original color information S of the long - focal - length / short - focal - length, the traffic - light fused color information C S and the long - focal - length / short - focal - length weights. Specifically, if the original target box K of the long - focal - length target image A1 has a higher coincidence degree with the fused target box C of the fused image C K and the original color information S of the long - focal - length target image A1 is the same as the fused color information C of the fused image C S then the final output color is based on the original color information S of the long - focal - length target image A1, and the position of the output target box is obtained by weighting the original target box K of the long - focal - length target image A1 and the fused target box C of the fused image C K according to the long - focal - length weight; if the original target box K of the long - focal - length target image A1 has a higher coincidence degree with the fused target box C of the fused image C K but the original color information S of the long - focal - length target image A1 is different from the fused color information C of the fused image C SIf they are different, the final output color is determined by the detection with higher confidence in the long - focal - length target image A1 and the short - focal - length target image B1. The position of the output target box is obtained by weighting the original target box K of the long - focal - length target image A1 and the original target box K of the short - focal - length target image B1 according to the long - focal - length weight. For example, if the fused target box C K has a greater degree of overlap with the original target box K of the long - focal - length image, and the original color information S of the long - focal - length image is the same as the traffic - light fused color information C S , then the output color is based on the original color information S of the long - focal - length image; moreover, the position information of the output target box is obtained by weighting the original target box K of the long - focal - length image and the fused target box C K , and the weight for weighting is based on the long - focal - length weight obtained in the above steps.

[0053] Finally, the output color and the output target box are packaged into a device result and sent to the vehicle navigation system to complete the traffic - light recognition.

[0054] As Figure 2 shown, in another embodiment, there is also a traffic - light recognition system based on a binocular camera. This system consists of five parts: an image acquisition module, an image fusion module, a weight calculation module, a data processing module, and a result sending module. The five modules cooperate with each other to execute the above - mentioned traffic - light recognition method. Through the acquisition of the image acquisition module, the fusion processing of the image data by the image fusion module, the division and recognition of the region of interest and the weight acquisition by the weight calculation module, and the calculation and acquisition of the output color and the output target box by the data processing module, finally, the result sending module performs the sending action, so that the vehicle navigation system receives the most accurate traffic - light recognition data, and then controls the autonomous driving action of the vehicle.

[0055] In another embodiment, a readable storage medium applicable to an in - vehicle computer is also disclosed, on which there is a readable storage medium for executing any one of the above - mentioned traffic - light recognition methods based on a binocular camera. This readable storage medium is a Flash storage chip or a high - speed solid - state drive, which has the advantages of large storage space, fast read - write speed, and strong stability.

[0056] Multiple embodiments of the present disclosure have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary technical personnel in the technical field to understand the disclosed embodiments.

Claims

1. A traffic light recognition method based on a binocular camera, characterized in that: The method is as follows: S01, setting up binocular cameras, wherein the binocular cameras are respectively a long-focus camera and a short-focus camera, and are respectively used to obtain a long-focus target image A1 and a short-focus target image B1 about a traffic light; extracting original color information S and original target frame K of the long-focus target image A1 and the short-focus target image B1; S02, fusing the long-focus target image A1 and the short-focus target image B1 to generate a fused image C; obtaining a fused target frame C of the fused image C K and fused color information C S ; S03, obtaining a telephoto region of interest ROI1 of the telephoto target image A1; Obtain a short-focus region of interest ROI2 of the short-focus target image B1; obtain long-focus / short-focus weights respectively according to the overlap degree between the original target frame K and the region of interest; S04, calculating the fusion target frame C K The degree of overlap with the original target frame K of the long / short focus, and according to the original color information S of the long / short focus and the fused color information C of the traffic light S and the long focus / short focus weight to determine the output color and the output target frame; if the original target frame K of the long focus target image A1 and the fused target frame C of the fused image C K The overlap is higher, and the original color information S of the telephoto target image A1 and the fused color information C of the fused image C S The final output color is based on the original color information S of the telephoto target image A1, and the position of the output target frame is determined by the original target frame K of the telephoto target image A1 and the fused target frame C of the fused image C. K According to the telephoto weight, if the original target frame K of the telephoto target image A1 and the fused target frame C of the fused image C K The degree of overlap is higher, but the original color information S of the telephoto target image A1 and the fused color information C of the fused image C S If they are different, the final output color is based on the detection with greater confidence in the long-focus target image A1 and the short-focus target image B1, and the position of the output target frame is obtained by weighting the original target frame K of the long-focus target image A1 and the original target frame K of the short-focus target image B1 according to the long-focus weight; S05. Pack the output color and the output target frame into a recognition package and send it to the vehicle navigation system to complete the traffic light recognition.

2. The traffic light recognition method based on binocular camera according to claim 1 is characterized in that: In step S02, before the long-focus target image A1 and the short-focus target image B1 are image-fused, an image preprocessing step is further included: obtaining a long-focus image pixel block according to the long-focus target image A1; and obtaining a short-focus image pixel block according to the short-focus target image B1.

3. The traffic light recognition method based on binocular camera according to claim 2 is characterized in that: In step S03, the long-focus region of interest ROI1 occupies 1 / 4 of the total pixel area of ​​the long-focus target image A1; the short-focus region of interest ROI2 occupies 1 / 5 of the total pixel area of ​​the short-focus target image B1.

4. The traffic light recognition method based on binocular camera according to claim 3 is characterized in that: In step S03, when calculating the overlap between the original target frame K of the long focus / short focus and the corresponding region of interest to obtain the long focus / short focus weights respectively, a normalization operation is also included.

5. The traffic light recognition method based on binocular camera according to claim 4 is characterized in that: In step S04, according to the original color information S of the long focus / short focus and the fused color information C of the traffic light, S When determining the output color, it also includes a confidence determination operation: when the original color information S of the long focus / short focus is combined with the fused color information C of the traffic light S When there is no match, the final traffic light color is set to the original color information S of the one with greater confidence in the long-focus / short-focus image.

6. The traffic light recognition system based on binocular camera collaboration is characterized by: The system includes: An image acquisition module, used to acquire a long-focus target image A1 and a short-focus target image B1 of a traffic light; and extract original color information S and original target frame K of the long-focus target image A1 and the short-focus target image B1 respectively; An image fusion module is used to fuse the long-focus target image A1 and the short-focus target image B1 to generate a fused image C; obtain a fused target frame C of the fused image C K and fused color information C S ; A weight calculation module, used for obtaining a telephoto region of interest ROI1 of the telephoto target image A1; Obtain a short-focus region of interest ROI2 of the short-focus target image B1; obtain long-focus / short-focus weights respectively according to the overlap degree between the original target frame K and the region of interest; Data processing module, used to calculate the fusion target box C K The degree of overlap with the original target frame K of the long / short focus, and according to the original color information S of the long / short focus and the fused color information C of the traffic light S and the long focus / short focus weight to determine the output color and the output target frame; if the original target frame K of the long focus target image A1 and the fused target frame C of the fused image C K The overlap is higher, and the original color information S of the telephoto target image A1 and the fused color information C of the fused image C S The final output color is based on the original color information S of the telephoto target image A1, and the position of the output target frame is determined by the original target frame K of the telephoto target image A1 and the fused target frame C of the fused image C. K According to the telephoto weight, if the original target frame K of the telephoto target image A1 and the fused target frame C of the fused image C K The degree of overlap is higher, but the original color information S of the telephoto target image A1 and the fused color information C of the fused image C S If they are different, the final output color is based on the detection with greater confidence in the long-focus target image A1 and the short-focus target image B1, and the position of the output target frame is obtained by weighting the original target frame K of the long-focus target image A1 and the original target frame K of the short-focus target image B1 according to the long-focus weight; The result sending module is used to package the output color and the output target frame into a recognition result and send it to the vehicle navigation system to complete the traffic light recognition.

7. The traffic light recognition system based on binocular camera collaboration according to claim 6 is characterized in that: The image fusion module further includes an image preprocessing module, which is used to obtain a long-focus image pixel block according to the long-focus target image A1; and obtain a short-focus image pixel block according to the short-focus target image B1.

8. The traffic light recognition system based on binocular camera collaboration according to claim 7 is characterized in that: The data processing module also includes a confidence determination module, which is used to determine the confidence when the original color information S of the long focus / short focus is combined with the fused color information C of the traffic light. S When there is no match, the final traffic light color is set to the original color information S of the one with greater confidence in the long-focus / short-focus image.

9. A readable storage medium suitable for an on-board computer, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, it is used to implement the traffic light recognition method based on a binocular camera as described in any one of claims 1 to 5.

Images