Traffic light status detection method and system based on single camera

Abstract

The present invention provides a traffic light state detection method and system based on a single camera, including: an image acquisition module acquires an original image and sets an area of ​​interest on the original image; an object detection module detects whether there is a traffic light in the area of ​​interest image; a state detection module detects the state of the traffic light in the area of ​​interest image and outputs the result; at the same time, it detects whether the traffic light is in the center of the area of ​​interest image, if so, the area of ​​interest parameters remain unchanged, if not, the parameter fine-tuning module adjusts the area of ​​interest parameters; it determines whether the cumulative number of images of the area of ​​interest in which the traffic light is not detected continuously exceeds a threshold, if not, the area of ​​interest parameters remain unchanged, if so, the area of ​​interest parameters are adjusted. The present invention uses a single high-resolution camera to replace the telephoto and short-focus cameras, reduces equipment costs, simplifies the detection algorithm, and the area of ​​interest sliding window method combining fine-tuning and coarse-tuning avoids the problem of increased model data volume caused by high-resolution images.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and in particular to a traffic light status detection method and system based on a single camera. Background Art

[0002] In autonomous driving technology, in order to ensure that autonomous driving vehicles can safely pass through intersections, autonomous driving vehicles are required to be able to detect the status of traffic lights at intersections in real time and identify red, green and yellow lights. After obtaining the status of these lights, they are output to the control system of the autonomous driving vehicle in the background, and the control system issues a pass or stop command based on the status of the lights.

[0003] In the prior art, when detecting traffic lights, autonomous vehicles perform collection and detection based on a single or multiple cameras. The existing single camera collection and detection of traffic lights has the problem of not being able to have both field of view angle and effective distance. The larger the field of view angle, the smaller the image of the same target, which increases the difficulty of image detection. If the field of view angle is smaller, the larger the image of the same target, the difficulty of image detection becomes lower, but the problem is that the camera field of view is small, and the traffic light can easily deviate from the field of view, causing the unmanned vehicle to be unable to detect the status of the traffic light ahead. However, detection with multiple cameras is not only costly, but also more complicated in detection logic, consuming more computing resources. Using automatic exposure at night will cause large halos and ghosting in the image, slower imaging speed, and the imaging of traffic lights is too different from that during the day, which is not conducive to effective detection.

[0004] Chinese patent CN110542931B discloses a "traffic light detection method, device, electronic device and computer readable medium", and its technical solution includes: real-time detection of the current position information of the vehicle; if it is determined that the vehicle has arrived at a preset intersection according to the current position information, the traffic light detection module is started and the traffic light at the preset intersection is detected based on deep learning; if the detection result of the traffic light detection module is that the traffic light is not detected, the angle information of the traffic light detection module is adjusted based on the step length of deep learning until the traffic light detection module detects the traffic light. According to the situation of different traffic light intersections, the method timely adjusts the angle information of the traffic light detection module when the traffic light is not detected until the traffic light detection module detects the traffic light, solving the problem that the automatic driving equipment cannot detect the status of the traffic light at some intersections. However, the above technical solution requires the vehicle to arrive at the preset intersection first and then perform the detection. When the traffic light cannot be detected at the first time, the detection module needs to be adjusted until the traffic light is detected. There is a problem that the detection time is too long, resulting in the vehicle needing to wait for a long time at the intersection, and there are certain safety hazards.

[0005] Chinese patent CN113947769A discloses a "traffic light detection method, device and storage medium based on long- and short-focus dual cameras", and its technical solution is: during the driving process of the vehicle, the first image captured by the long-focus camera is sequentially subjected to traffic light detection to obtain a first bounding box; the first bounding box is transformed into the pixel coordinate system of the short-focus camera to obtain a second bounding box, and the position of the second bounding box at the next moment is predicted in the pixel coordinate system of the short-focus camera; when the traffic light is not detected in the first image for the first time, the second image captured by the short-focus camera is subjected to traffic light detection to obtain a third bounding box of the traffic light in the second image; if the third bounding box matches the predicted position of the second bounding box at the current moment, it is determined that the third bounding box detects the traffic light. Its existence is costly, and the detection logic is complex, consuming more computing resources.

[0006] The target is detected by using multiple cameras (long-focus and short-focus). If the long-focus camera can see clearly, the long-focus camera is used for detection. If the long-focus camera is not clear, the short-focus (wide-angle) camera is used for detection. Usually, a 2-megapixel or 5-megapixel camera is used for detection, and the automatic exposure mode is used.

[0007] Using two or more cameras for detection is not only costly, but also consumes more computing resources and makes the detection logic more complicated. Using automatic exposure at night will cause large halos and smears in the image, and the imaging speed will be slower. In addition, the imaging of traffic lights is too different from that during the day, which is not conducive to the detection module to perform effective detection.

[0008] The existing solution uses multiple cameras, which is costly and has high computing resource utilization. Multiple cameras make the algorithm logic more complicated. The smear and halo problems caused by automatic exposure lead to poor imaging quality of traffic lights at night, making detection more difficult.

[0009] Therefore, it is necessary to propose a technical solution to improve the above technical problems. Summary of the invention

[0010] In view of the defects in the prior art, an object of the present invention is to provide a traffic light status detection method and system based on a single camera.

[0011] A traffic light status detection method based on a single camera provided by the present invention includes:

[0012] Step S1: The image acquisition module acquires an original image of the road environment, and then sets an area of ​​interest on the original image according to the area of ​​interest parameters, and captures an image of the area of ​​interest;

[0013] Step S2: the target detection module detects whether there is a traffic light in the image of the region of interest. If so, the process proceeds to step S3; otherwise, the process jumps to step S4;

[0014] Step S3: the state detection module detects the state of the traffic light in the image of the region of interest, determines whether it is a red light, a yellow light or a green light, and outputs the result; at the same time, the target detection module is used to detect whether the traffic light is in the center of the image of the region of interest. If so, the parameters of the region of interest remain unchanged; if not, the parameters of the region of interest are finely adjusted by the parameter fine-tuning module;

[0015] Step S4: determine whether the cumulative number of frames of the ROI image in which the traffic lights are not detected continuously exceeds a preset threshold. If not, the ROI parameters remain unchanged. If so, the ROI parameters are substantially adjusted using the parameter coarse adjustment module.

[0016] Preferably, the original image and the ROI are both rectangular, the size of the ROI is smaller than that of the original image, and the ROI image is a part of the original image; the ROI parameters include the size of the ROI and the position of the center of the ROI, both of which are set to initial values.

[0017] Preferably, the target detection module and the state detection module use deep learning networks or traditional computer vision technology to detect the image of the region of interest; the target detection module and the state detection module can be the same module or two independent modules.

[0018] Preferably, the fine adjustment of the parameters of the region of interest by means of a parameter fine adjustment module comprises:

[0019] According to the position of the traffic light in the image of the region of interest, the position of the traffic light in the original image is obtained;

[0020] Adjust the position of the center of the region of interest to the position of the traffic light in the original image;

[0021] If the fine-tuned ROI exceeds the range of the original image, the size of the ROI is reduced accordingly so that the ROI does not exceed the boundary of the original image, or the size of the ROI remains unchanged and the position of the center of interest is adjusted to a position closest to the position of the traffic light in the original image so that the ROI does not exceed the boundary of the original image; if the fine-tuned ROI does not exceed the range of the original image, if the size of the ROI is smaller than the initial value at this time, the size of the ROI is expanded outward to the lower limit of the initial value and the boundary of the original image.

[0022] Preferably, the substantially adjusting the parameters of the region of interest by means of a coarse parameter adjustment module comprises:

[0023] The original image is divided into N small blocks of the same size and overlapped with each other, and the size of the region of interest is set to the size of the small block;

[0024] Set the center of the region of interest to the center of the first small block;

[0025] The region of interest is set on the original image according to the above-mentioned region of interest parameters, and the region of interest image is obtained by intercepting;

[0026] The target detection module detects whether there is a traffic light in the image of the ROI. If not, the center of the ROI is set to the center of the second small block, and so on, until the Nth small block; if it exists, the size of the ROI is adjusted to the initial value and the coarse parameter adjustment is ended.

[0027] Preferably, the ROI parameters are finely or roughly adjusted and then used for interception and detection of the next frame of original image.

[0028] The present invention also provides a traffic light status detection system based on a single camera, comprising:

[0029] Image acquisition module: used to collect original images of the road environment;

[0030] Region of interest setting module: sets the region of interest on the original image according to the region of interest parameters, and obtains the region of interest image;

[0031] Object detection module: used to detect whether there is a traffic light in the image of the region of interest, and whether the traffic light is in the center of the image of the region of interest;

[0032] State detection module: used to detect the state of the traffic light in the image of the area of ​​interest, determine whether it is red, yellow or green, and output the result;

[0033] Parameter fine-tuning module: used to fine-tune the parameters of the region of interest so that the traffic light is located in the center of the region of interest, thereby enhancing the robustness of detection;

[0034] Parameter coarse adjustment module: used to significantly adjust the parameters of the region of interest and slide the region of interest on the original image to increase the probability of detecting the traffic light.

[0035] Preferably, the image acquisition module uses a single large-resolution camera and adopts a fixed exposure time.

[0036] Preferably, the target detection module and the state detection module use deep learning networks or traditional computer vision technology to detect the image of the region of interest; the target detection module and the state detection module can be the same module or two independent modules.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. The present invention uses a single high-resolution camera to replace the long- and short-focus cameras, which reduces the equipment cost and simplifies the algorithm logic for traffic light detection;

[0039] 2. The present invention adopts a fixed exposure time instead of the traditional automatic exposure. Since the traffic light is an active luminous target and the luminous brightness is basically constant, the image acquisition can be performed by adopting a fixed exposure time, which ensures the consistency of imaging effects during the day and at night, avoids the problems of smear and halo caused by too long automatic exposure time at night, and makes the traffic light easier to detect;

[0040] 3. Since the use of large-resolution images directly sent to the deep learning network for detection will increase the detection time of the model and cannot meet the real-time requirements, the present invention designs a sliding window method of the region of interest that combines fine adjustment and coarse adjustment, which ensures the detection effect of traffic lights without increasing the calculation amount of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0042] Figure 1 It is the overall logic diagram of the present invention;

[0043] Figure 2 It is a logic diagram of the parameter fine-tuning module of the present invention;

[0044] Figure 3 This is a logic diagram of the parameter coarse adjustment module of the present invention. DETAILED DESCRIPTION

[0045] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0046] Embodiment 1:

[0047] like Figure 1 As shown, the present invention provides a traffic light state detection method based on a single camera, comprising:

[0048] Step S1: The image acquisition module acquires an original image of the road environment, and then sets an area of ​​interest on the original image according to the area of ​​interest parameters, and captures an image of the area of ​​interest;

[0049] Step S2: The target detection module detects whether there is a traffic light in the image of the region of interest. If so, the process proceeds to step S3, otherwise, the process jumps to step S4;

[0050] Step S3: The state detection module detects the state of the traffic light in the image of the region of interest, determines whether it is a red light, a yellow light or a green light, and outputs the result; at the same time, the target detection module is used to detect whether the traffic light is in the center of the image of the region of interest. If so, the parameters of the region of interest remain unchanged. If not, the parameters of the region of interest are finely adjusted using the parameter fine-tuning module;

[0051] Step S4: determine whether the cumulative number of frames of the ROI image in which the traffic lights are not detected continuously exceeds a preset threshold. If not, the ROI parameters remain unchanged. If so, the ROI parameters are substantially adjusted using the parameter coarse adjustment module.

[0052] Furthermore, the original image and the region of interest are both rectangular, the size of the region of interest is smaller than the size of the original image, and the region of interest image is part of the original image; thereby reducing the amount of data and time spent on a single detection. The region of interest should not be too large or too small. If it is too large, the data volume will not be significantly reduced; if it is too small, the number of detections will increase, which will reduce efficiency; the region of interest parameters include the size of the region of interest and the position of the center of the region of interest, both of which are set to initial values.

[0053] Furthermore, the target detection module and the state detection module use deep learning networks or traditional computer vision technology to detect the image of the region of interest; the target detection module and the state detection module can be the same module or two independent modules.

[0054] Further, such as Figure 2 As shown, the parameters of the region of interest are finely adjusted through the parameter fine-tuning module, including:

[0055] According to the position of the traffic light in the image of the region of interest, the position of the traffic light in the original image is obtained;

[0056] Adjust the position of the center of the region of interest to the position of the traffic light in the original image;

[0057] If the fine-tuned ROI exceeds the range of the original image, the size of the ROI is reduced accordingly so that the ROI does not exceed the boundary of the original image, or the size of the ROI remains unchanged and the position of the center of interest is adjusted to a position closest to the position of the traffic light in the original image so that the ROI does not exceed the boundary of the original image; if the fine-tuned ROI does not exceed the range of the original image, if the size of the ROI is smaller than the initial value at this time, the size of the ROI is expanded outward to the lower limit of the initial value and the boundary of the original image.

[0058] Further, such as Figure 3 As shown, the parameters of the region of interest are substantially adjusted through the parameter coarse adjustment module, including:

[0059] The original image is divided into N small blocks of the same size and overlapped with each other, and the size of the region of interest is set to the size of the small block;

[0060] Set the center of the region of interest to the center of the first small block;

[0061] The region of interest is set on the original image according to the above-mentioned region of interest parameters, and the region of interest image is obtained by intercepting;

[0062] The target detection module detects whether there is a traffic light in the image of the ROI. If not, the center of the ROI is set to the center of the second small block, and so on, until the Nth small block; if it exists, the size of the ROI is adjusted to the initial value and the coarse parameter adjustment is ended.

[0063] Furthermore, after fine-tuning or coarse-tuning, the parameters of the region of interest are used for interception and detection of the next frame of the original image.

[0064] Embodiment 2:

[0065] Compared with Example 1, this example describes the present invention in more detail.

[0066] The present invention also provides a traffic light status detection system based on a single camera, comprising:

[0067] Image acquisition module: used to collect original images of the road environment;

[0068] Region of interest setting module: sets the region of interest on the original image according to the region of interest parameters, and obtains the region of interest image;

[0069] Object detection module: used to detect whether there is a traffic light in the image of the area of ​​interest, and whether the traffic light is in the center of the image of the area of ​​interest;

[0070] State detection module: used to detect the state of the traffic light in the image of the area of ​​interest, determine whether it is red, yellow or green, and output the result;

[0071] Parameter fine-tuning module: used to fine-tune the parameters of the region of interest so that the traffic light is located in the center of the region of interest, thereby enhancing the robustness of detection;

[0072] Parameter coarse adjustment module: used to significantly adjust the parameters of the region of interest and slide the region of interest on the original image to increase the probability of detecting the traffic light.

[0073] Furthermore, the image acquisition module uses a single large-resolution camera and adopts a fixed exposure time.

[0074] Furthermore, the target detection module and the state detection module use deep learning networks or traditional computer vision technology to detect the image of the region of interest; the target detection module and the state detection module can be the same module or two independent modules.

[0075] The present invention uses a single high-resolution camera to replace multiple long- and short-focus cameras in the prior art, thereby reducing the investment cost of equipment. Moreover, due to the reduction in cameras, the algorithm logic of traffic light detection is simplified, thereby reducing the calculation cost.

[0076] Embodiment 3:

[0077] Embodiment 3 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0078] The present invention provides a traffic light state detection method and system based on a single camera, including an image acquisition module, an area of ​​interest setting module, a target detection module, a state detection module, a parameter fine adjustment module and a parameter coarse adjustment module. The detection method is as follows Figure 1 As shown, the following steps are included:

[0079] Step S1: the image acquisition module acquires the original image of the road environment, and the region of interest setting module sets the region of interest on the original image according to the region of interest parameters, and captures the region of interest image;

[0080] Step S2: The target detection module detects whether there is a traffic light in the image of the region of interest. If so, the process proceeds to step S3; otherwise, the process jumps to step S4;

[0081] Step S3: the state detection module detects the state of the traffic light in the image of the region of interest, determines whether it is a red light, a yellow light or a green light, and outputs the result; at the same time, the target detection module is used to detect whether the traffic light is in the center of the image of the region of interest. If so, the parameters of the region of interest remain unchanged; if not, the parameters of the region of interest are finely adjusted by the parameter fine-tuning module;

[0082] Step S4: determine whether the cumulative number of frames of the ROI image in which the traffic lights are not detected continuously exceeds a preset threshold. If not, the ROI parameters remain unchanged. If so, the ROI parameters are substantially adjusted using the parameter coarse adjustment module.

[0083] In this embodiment, the following preferred solution is adopted:

[0084] Image acquisition module:

[0085] In this embodiment, a 12-megapixel, 8mm focal length, high-resolution, large FOV (field of view) camera is used, wherein the camera FOV is greater than 100 degrees (the focal length of the ultra-wide-angle lens is 20-13 mm, and the viewing angle is 94-118 degrees), and the image resolution is 4024x3036 pixels. This camera has the advantages of both telephoto and wide-angle cameras.

[0086] Since traffic lights are actively luminous targets and their brightness is basically constant, images can be captured using a fixed exposure time. Fixed exposure time ensures consistency in imaging effects during the day and at night, avoids problems such as smearing and halos caused by long automatic exposure time at night, and makes traffic lights easier to detect.

[0087] Region of interest setting module:

[0088] Since the original image is relatively large, in order to ensure the real-time detection, the region of interest is captured for detection by setting the region of interest parameters.

[0089] In this embodiment, the parameters of the region of interest include size and center position, the initial value is the center of the original image, and the size is in the range of 1920×1080.

[0090] Object detection module:

[0091] It uses a deep learning network (such as faster-rcnn, yolov5, centernet, etc.) target detection network to detect whether there is a traffic light in the input image of interest, and whether the traffic light is in the center of the image of the area of ​​interest.

[0092] Status detection module:

[0093] It is used to detect the status of traffic lights in the image of the area of ​​interest, determine whether it is red, yellow or green, and output the result.

[0094] Parameter fine-tuning module:

[0095] It is used to fine-tune the parameters of the region of interest so that the traffic light is located in the center of the region of interest, thereby enhancing the robustness of detection.

[0096] Parameter coarse adjustment module:

[0097] It is used to significantly adjust the parameters of the region of interest and slide the region of interest on the original image to increase the probability of detecting the traffic light.

[0098] In this example, the parameter coarse adjustment module divides the original image into three small blocks of the same size and overlap, and the region of interest setting module sets the size of the region of interest to the size of this small block. When the target detection module fails to detect the traffic light for a long time, the cumulative number of frames of the region of interest image exceeds the preset threshold. If not, the parameter coarse adjustment module is started (if it is less than the threshold, it is considered that it may be caused by missed detection), and the region of interest image is slid on the three areas in turn until the traffic light can be detected. The purpose is to find the image area where the target may exist and increase the probability of detecting the traffic light.

[0099] Since the use of large-resolution images directly sent to the deep learning network for detection will increase the detection time of the model and cannot meet the real-time requirements, the present invention designs a region of interest sliding window method that combines fine adjustment and coarse adjustment, which ensures the detection effect of traffic lights without increasing the calculation amount of the model.

[0100] In this embodiment, all camera parameters in the image acquisition module can be changed according to actual conditions, and are not limited to 12 million pixels, 8mm focal length, 4024x3036 pixel resolution, 2ms fixed exposure time and initial region of interest parameters of 1920x1080 range.

[0101] The parameter coarse adjustment module is not limited to using three small blocks of the same size and overlapped with each other, and can be changed according to the actual situation. The target detection module and the state detection module are not limited to using deep learning networks for detection, but can also be traditional computer vision solutions, and the two can be the same integrated module or two independent modules.

[0102] Those skilled in the art may understand this embodiment as a more specific description of Embodiment 1 and Embodiment 2.

[0103] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0104] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A traffic light status detection method based on a single camera, characterized in that: include: Step S1: The image acquisition module acquires an original image of the road environment, and then sets an area of ​​interest on the original image according to the area of ​​interest parameters, and captures an image of the area of ​​interest; Step S2: the target detection module detects whether there is a traffic light in the image of the region of interest. If so, the process proceeds to step S3; otherwise, the process jumps to step S4; Step S3: the state detection module detects the state of the traffic light in the image of the region of interest, determines whether it is a red light, a yellow light or a green light, and outputs the result; at the same time, the target detection module is used to detect whether the traffic light is in the center of the image of the region of interest. If so, the parameters of the region of interest remain unchanged; if not, the parameters of the region of interest are finely adjusted by the parameter fine-tuning module; Step S4: determining whether the cumulative number of frames of the ROI images in which the traffic lights are not detected continuously exceeds a preset threshold; if not, the parameters of the ROI remain unchanged; if yes, the parameters of the ROI are substantially adjusted using a parameter coarse adjustment module; The original image and the region of interest are both rectangular, the size of the region of interest is smaller than the size of the original image, and the region of interest image is a part of the original image; the region of interest parameters include the size of the region of interest and the position of the center of the region of interest, both of which are set to initial values; The target detection module and the state detection module detect the image of the region of interest using a deep learning network or traditional computer vision technology; the target detection module and the state detection module may be the same module or two independent modules; The fine adjustment of the parameters of the region of interest by the parameter fine adjustment module includes: According to the position of the traffic light in the image of the region of interest, the position of the traffic light in the original image is obtained; Adjust the position of the center of the region of interest to the position of the traffic light in the original image; If the fine-tuned ROI exceeds the range of the original image, the size of the ROI is reduced accordingly so that the ROI does not exceed the boundary of the original image, or the size of the ROI remains unchanged and the position of the center of interest is adjusted to a position closest to the position of the traffic light in the original image so that the ROI does not exceed the boundary of the original image; if the fine-tuned ROI does not exceed the range of the original image, if the size of the ROI is smaller than the initial value at this time, the size of the ROI is expanded outward to the lower limit of the initial value and the boundary of the original image; The step of substantially adjusting the parameters of the region of interest by means of a coarse parameter adjustment module includes: Divide the original image into N There are small blocks of the same size and overlapped with each other, and the size of the region of interest is set to the size of the small block; Set the center of the region of interest to the center of the first small block; The region of interest is set on the original image according to the above-mentioned region of interest parameters, and the region of interest image is obtained by intercepting; The target detection module detects whether there is a traffic light in the image of the region of interest. If not, the center of the region of interest is set to the center of the second small block, and so on until the N If there is a small block, adjust the size of the region of interest to the initial value and end the coarse adjustment of parameters; After fine adjustment or rough adjustment, the parameters of the region of interest are used for interception and detection of the next frame of original image.

2. A traffic light status detection system based on a single camera, characterized in that: The system is used to execute the traffic light state detection method based on a single camera according to claim 1, comprising: Image acquisition module: used to collect original images of the road environment; Region of interest setting module: sets the region of interest on the original image according to the region of interest parameters, and obtains the region of interest image; Object detection module: used to detect whether there is a traffic light in the image of the region of interest, and whether the traffic light is in the center of the image of the region of interest; State detection module: used to detect the state of the traffic light in the image of the area of ​​interest, determine whether it is red, yellow or green, and output the result; Parameter fine-tuning module: used to fine-tune the parameters of the region of interest so that the traffic light is located in the center of the region of interest, thereby enhancing the robustness of detection; Parameter coarse adjustment module: used to significantly adjust the parameters of the region of interest and slide the region of interest on the original image to increase the probability of detecting the traffic light.

3. The traffic light status detection system based on a single camera according to claim 2 is characterized in that: The image acquisition module uses a single large-resolution camera and adopts a fixed exposure time.

4. The traffic light status detection system based on a single camera according to claim 2 is characterized in that: The target detection module and the state detection module use deep learning networks or traditional computer vision technology to detect the image of the region of interest; the target detection module and the state detection module can be the same module or two independent modules.

Images