Dynamic traffic signal lamp detection method and device, equipment and storage medium

Abstract

The invention provides a dynamic traffic signal lamp detection method and device, equipment and a storage medium. According to the method, traffic signal lamp detection is realized through combination of a high-precision MCD algorithm and continuous frame time correlation constraints, and a multi-stage feature fusion and dynamic parameter adjustment mechanism is adopted, so that the problem of low signal lamp recognition accuracy in a complex dynamic environment is effectively solved; the method has the remarkable advantages of improving the detection precision, enhancing the environmental adaptability and guaranteeing the safety of automatic driving.

Description

Technical Field

[0001] This disclosure relates to the field of autonomous driving, and in particular to a method, apparatus, device, and storage medium for detecting dynamic traffic lights. Background Technology

[0002] In the fields of intelligent transportation and autonomous driving, high-precision maps serve as the core foundation for decision-making in autonomous vehicles, playing a crucial role in the accurate identification and timely updating of traffic lights. Currently widely used map change detection (MCD) systems primarily identify newly added, missing, or abnormally positioned map elements by comparing onboard perception data with map data.

[0003] However, current detection methods suffer from low recognition accuracy in complex intersection scenarios due to the complex environment and the relatively increased number of traffic lights, leading to false alarms and missed alarms. Summary of the Invention

[0004] In view of this, the purpose of this disclosure is to provide a dynamic traffic light detection method, device, equipment and storage medium based on a high-precision MCD algorithm, so as to solve the problems of low accuracy of existing traffic light detection and inability to cope with the frequent changes of complex intersections, resulting in detection lag, missed detection and map update delay.

[0005] In a first aspect, embodiments of this disclosure provide a dynamic traffic light detection method based on a high-precision MCD algorithm, including: A continuous sequence of images of the traffic environment is captured in real time by a high-resolution camera mounted on the vehicle, wherein the continuous sequence of images includes dynamically changing traffic light targets; The continuous image sequence is subjected to image denoising, illumination equalization and image enhancement processing to obtain an enhanced image sequence; The enhanced image sequence is used to detect and locate traffic light targets based on the high-precision MCD algorithm, and the detection and localization results are constrained by the temporal correlation between consecutive frames to obtain traffic light candidate regions. The candidate regions for traffic lights are subjected to color and shape feature extraction. Based on the extracted features, the status of the traffic lights is determined, and the detection results of the traffic lights are output. The detection results include determining the color status and direction status of each traffic light.

[0006] Secondly, embodiments of this disclosure provide a dynamic traffic light detection device based on a high-precision MCD algorithm, comprising: The acquisition module is used to acquire a continuous sequence of images in the traffic environment in real time using a high-resolution camera installed on the vehicle, wherein the continuous sequence of images includes dynamically changing traffic light targets; The enhancement module is used to perform image denoising, illumination equalization and image enhancement processing on the continuous image sequence to obtain an enhanced image sequence; The detection module is used to detect and locate traffic light targets in the enhanced image sequence based on the high-precision MCD algorithm, and to constrain the detection and localization results by utilizing the temporal correlation constraints between consecutive frames to obtain traffic light candidate regions. The determination module is used to extract color and shape features from the candidate areas of the traffic lights, determine the state of the traffic lights based on the extracted features, and output the detection results of the traffic lights. The detection results include determining the color state and direction state of each traffic light.

[0007] Thirdly, embodiments of this disclosure provide an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-described dynamic traffic light detection method based on a high-precision MCD algorithm.

[0008] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to implement the aforementioned dynamic traffic light detection method based on a high-precision MCD algorithm.

[0009] The embodiments disclosed herein bring the following beneficial effects: The aforementioned dynamic traffic light detection method, device, equipment, and storage medium, through real-time acquisition of continuous image sequences by a high-resolution camera, can capture the complete process of dynamic changes in traffic lights, providing a continuous data foundation in the time dimension for subsequent analysis. Secondly, image denoising, illumination equalization, and enhancement processing effectively eliminate the influence of environmental noise and illumination deviations, improving the distinction between the traffic light target and the background, and providing high-quality input data for the MCD algorithm. Furthermore, based on a high-precision MCD algorithm combined with continuous frame temporal correlation constraints, it can filter out single-frame false detections through multi-frame information fusion and capture the continuity of traffic light movement, reducing the risk of missed detections. Finally, through the extraction of color and shape features and multi-frame state fusion judgment, accurate identification of the color and direction status of traffic lights is achieved, avoiding misjudgments caused by single-frame flickering or viewpoint deviations. In summary, the synergistic effect of the above technical features improves the detection accuracy of the method at complex intersections compared to traditional MCD algorithms, effectively solving the problems of detection lag, missed detections, and map update delays in existing technologies.

[0010] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objects and other advantages of this disclosure are realized and obtained through the structures particularly pointed out in the description, claims and drawings.

[0011] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the specific embodiments of this disclosure or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 This is a flowchart of one embodiment of the dynamic traffic light detection method based on the high-precision MCD algorithm in this disclosure; Figure 2 A schematic diagram of a dynamic traffic light detection device based on a high-precision MCD algorithm provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0015] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] In related technologies, intelligent transportation and autonomous driving rely on high-precision maps for decision-making, with the identification and updating of traffic lights directly impacting system reliability. Traditional map change detection systems identify abnormal elements by comparing onboard perception data with map data. However, in complex intersection scenarios, due to the large number of traffic lights and frequent dynamic changes in the environment, existing methods are susceptible to uneven lighting, motion blur, and background interference, leading to decreased detection accuracy and frequent false alarms and missed alarms. For example, in rainy, foggy, or backlit conditions, traditional algorithms struggle to accurately distinguish traffic lights from objects of similar color, and single-frame detection lacks temporal information support, making it difficult to handle target tracking problems in fast-moving scenarios.

[0017] To address the aforementioned issues, the inventors discovered that existing technologies fail to adequately utilize image preprocessing and temporal information, resulting in poor robustness in traffic light detection under complex scenarios. Analysis revealed that the key to improving detection accuracy lies in optimizing image quality and introducing temporal correlation constraints. Therefore, they proposed strengthening denoising and illumination equalization during the preprocessing stage, while simultaneously incorporating multi-frame information for dynamic correction during detection, thereby enhancing the algorithm's adaptability to complex environments.

[0018] For ease of understanding, the specific process of the embodiments of this disclosure is described below. The embodiments of this disclosure are applied to autonomous driving systems. Please refer to [link / reference]. Figure 1 One embodiment of the dynamic traffic light detection method based on the high-precision MCD algorithm in this disclosure includes: S10. A continuous sequence of images of the traffic environment is acquired in real time by a high-resolution camera installed on the vehicle, including dynamically changing traffic light targets. It should be noted that a high-resolution camera refers to an image acquisition device with high dynamic range imaging capabilities. Specifically, it can be achieved by combining a global shutter sensor with a fisheye lens, used to capture distant targets and suppress motion blur.

[0019] In this step, the vehicle performs autonomous driving based on a pre-planned route or a pre-updated high-precision map, and locates its own position in real time. When it determines that it is close to an intersection based on its location, it activates the camera at the front of the vehicle to capture images of the intersection ahead, including traffic light information.

[0020] Specifically, when the vehicle approaches an intersection, the operating parameters (exposure, focal length, etc.) of the onboard high-resolution camera (equipped with a fisheye lens, global shutter, and HDR function) are adjusted based on environmental parameters (such as lighting and vehicle speed). Simultaneously, continuous dynamic images are acquired, and combined with GPS positioning and inertial measurement unit (IMU) data, a continuous image sequence is generated by aligning the images according to timestamps. The camera supports forward / reverse multi-view acquisition and uses a polarizing filter to suppress glare, ensuring image quality under complex lighting conditions.

[0021] A panoramic camera equipped with a fisheye lens is used to cover a wider field of view to obtain traffic light images from a greater distance. The camera has a global shutter function to prevent image shaking and distortion when vehicles are moving at high speeds. It also supports high dynamic range imaging by fusing images from multiple exposures to cope with strong light and low light scenes, thereby improving detection accuracy under different brightness conditions.

[0022] Fisheye lenses refer to ultra-wide-angle optical lenses with a field of view exceeding 180 degrees. This can be achieved through a combination of multiple aspherical lens elements, increasing horizontal coverage to capture traffic lights at distant points in intersections. Global shutter functionality involves synchronously controlling the exposure time of all pixels on the image sensor, typically achieved using a CMOS sensor global shutter mode, to eliminate motion blur caused by high-speed vehicle movement. High dynamic range imaging involves continuously acquiring and fusing images with different exposure times, typically using a three-exposure stacking algorithm, to enhance image detail under extreme lighting conditions.

[0023] Specifically, the panoramic camera expands the horizontal field of view to 270 degrees by utilizing the optical distortion characteristics of a fisheye lens, enabling vehicles to capture traffic lights located on curves or at a distance as they approach intersections. The global shutter mode avoids the tilting distortion phenomenon caused by traditional rolling shutters when photographing moving objects by uniformly controlling the charge accumulation time of sensor pixels. High dynamic range imaging technology employs a three-exposure sequence (short, medium, and long) to preserve structural details in bright areas and brightness information in low-light areas, respectively. After weighted fusion, a feature image containing complete brightness levels is generated.

[0024] S20. Perform image denoising, illumination equalization and image enhancement processing on the continuous image sequence to obtain an enhanced image sequence; Understandably, this image denoising process refers to eliminating image noise through filtering algorithms. Specifically, a combination of Gaussian filtering and median filtering can be used to effectively remove random noise while preserving edge details.

[0025] This illumination equalization refers to adjusting the brightness distribution of an image. Specifically, it can be achieved using an adaptive histogram equalization method to improve details in dark areas and prevent overexposure.

[0026] In this step, when performing multi-dimensional enhancement processing on the continuous image sequence, the specific steps include: performing color space conversion (such as RGB to HSV) on each image in the continuous image sequence, denoising with a combination of Gaussian filtering and median filtering to preserve edge details, and then enhancing the contrast between the traffic lights and the background through adaptive histogram equalization and Retinex color restoration; cropping the road region (ROI) based on lane line detection, and initially extracting potential traffic light regions by combining Canny edge detection and color thresholding; and constructing a multi-scale Gaussian pyramid to take into account the detection needs of traffic lights at different distances.

[0027] Step S30: Detect and locate traffic light targets in the enhanced image sequence based on the high-precision MCD algorithm, and constrain the detection and localization results by utilizing the temporal correlation constraints between consecutive frames to obtain traffic light candidate regions; It is understood that the high-precision MCD algorithm refers to a detection model that integrates color histogram and shape template matching. Specifically, it can use a lightweight convolutional neural network to extract features and filter target regions by confidence calculation.

[0028] The aforementioned temporal correlation constraint refers to using multi-frame information for trajectory prediction, specifically by combining Kalman filtering with optical flow tracking to improve the temporal continuity of the detection results.

[0029] In this embodiment, when detecting, locating, and constraining traffic light targets in each image of the enhanced image sequence, the enhanced image is divided into grid regions, and potential candidate regions are screened based on the characteristics of traffic light color (red / green / yellow) and shape (circle, arrow, etc.). The pixel features of the candidate regions are extracted by a lightweight convolutional neural network (CNN) of the vehicle edge computing unit, and the confidence of color histogram and shape template matching is calculated by combining a high-precision MCD algorithm. The position of the traffic light in the previous frame is tracked using optical flow, and new regions are detected by combining background subtraction. The trajectory is predicted and position fluctuations are smoothed by Kalman filtering. Non-maximum suppression (NMS) is applied to eliminate overlapping detection boxes, and regions with consistent features across multiple frames are clustered to output high-confidence traffic light candidate regions.

[0030] In this step, optical flow is first used to track the traffic light targets detected in the previous frame to predict their position in the current frame; in the new areas not covered by optical flow tracking in the current frame, a motion detection algorithm based on background subtraction is used to identify moving targets; color features in the color space are extracted from the detected candidate areas, and these features are evaluated to see if they conform to the typical pattern of traffic lights.

[0031] Furthermore, shape features are extracted from the candidate traffic light regions, including the aspect ratio, roundness, and spacing between light groups. The extracted color, shape, and brightness features are then input into the trained MCD algorithm model, which calculates the confidence level of the candidate region as a traffic light. Based on the confidence level threshold, the candidate regions are classified, and low-confidence candidate regions are filtered out to reduce false detections.

[0032] Furthermore, each candidate region is continuously tracked across multiple frames, and the motion trajectory of the traffic lights is smoothly predicted using Kalman filtering. Non-maximum suppression (NMS) is applied to the candidate detection results to eliminate redundant detection boxes with overlapping positions. The detection results of the candidate regions are verified by utilizing the spatial consistency between adjacent pixels and the known traffic light structure (e.g., three-light arrangement), thereby further improving the accuracy.

[0033] Step S40: Extract color and shape features from the candidate areas of traffic lights, determine the state of the traffic lights based on the extracted features, and output the detection results of the traffic lights, which include determining the color state and direction state of each traffic light.

[0034] In this embodiment, color and shape features are extracted from the candidate regions, and the detection results are output by combining multi-frame state fusion. Specifically, this includes: First, HSV color space conversion and edge detection are performed to extract saturation, brightness and shape contours. Then, a color stability filter (statistical fluctuation range) is applied to screen the reliable region. Combined with shape closure, symmetry analysis and direction template matching, the color (red / yellow / green) and direction (straight, left turn, etc.) status of the traffic lights are determined. Finally, the states of multiple consecutive frames are fused using a sliding time window, written into a buffer queue, and the final detection results (color, direction, and position) are output.

[0035] Specifically, this embodiment first acquires dynamic image sequences using an optimized camera to ensure the quality of the original data. In the preprocessing stage, environmental noise is eliminated through color space conversion and multi-level filtering, followed by adaptive histogram equalization to enhance the contrast of the target area. In the detection stage, the enhanced image is divided into grid regions, and candidate regions are initially screened using color histogram matching, then further verified using shape templates. The position of the detection box is corrected using motion trajectory prediction between consecutive frames, and overlapping areas are eliminated through non-maximum suppression. Finally, HSV color space analysis and edge detection are performed on the candidate regions, and the traffic light status is determined by combining predefined discrimination rules. A sliding window mechanism is used to fuse multi-frame results to output stable detection data.

[0036] Compared to existing technologies, which typically employ single-frame detection and lack effective preprocessing, this solution significantly improves feature discriminability in low-light scenes through multi-level image enhancement. Traditional algorithms rely on fixed threshold segmentation, which is susceptible to environmental interference. This solution, by introducing temporal correlation constraints, can effectively distinguish static interference objects from real traffic lights by utilizing motion continuity. Furthermore, while existing technologies often use single color features for judgment, this solution combines shape template matching and direction analysis to accurately identify special shapes such as arrow traffic lights.

[0037] The above technical solution utilizes a high-precision MCD algorithm combined with continuous frame temporal correlation constraints to detect traffic lights. By employing a multi-stage feature fusion and dynamic parameter adjustment mechanism, it effectively solves the problem of low accuracy in traffic light recognition under complex dynamic environments. This approach offers significant advantages in improving detection accuracy, enhancing environmental adaptability, and ensuring the safety of autonomous driving.

[0038] Next, we will explain in detail each step of the dynamic traffic light detection method based on the high-precision MCD algorithm.

[0039] In one embodiment, the real-time acquisition of a continuous image sequence in the traffic environment using a high-resolution camera mounted on the vehicle includes: analyzing environmental parameters of the vehicle's location when the vehicle approaches an intersection, and adjusting the operating parameters of the camera mounted on the front of the vehicle based on the environmental parameters, wherein the operating parameters include at least one of exposure parameters and distance; acquiring continuous dynamic images of the intersection in real time using the adjusted camera, as well as recording the vehicle's GPS positioning information and attitude data measured by the inertial measurement unit, and aligning the GPS positioning information, attitude data, and continuous dynamic images according to timestamps to obtain a continuous image sequence.

[0040] The environmental parameters refer to the real-time light intensity, weather conditions, and road structure at the vehicle's location. Specifically, they can be implemented using a photosensitive sensor, a brightness analysis module built into the camera, and navigation map data to dynamically perceive changes in the intersection environment and optimize camera parameters.

[0041] The exposure parameters refer to the duration and sensitivity of the camera's photosensitive element in receiving light. Specifically, an automatic exposure algorithm can be used to dynamically adjust the exposure time based on the ambient brightness. For example, in a bright light environment, the exposure time can be shortened to avoid overexposure, while in a low light environment, the exposure time can be extended to enhance detail capture.

[0042] The GPS positioning information refers to the vehicle's latitude and longitude coordinates obtained through the Global Positioning System. Specifically, it can be achieved by using an onboard GPS module combined with differential positioning technology to achieve centimeter-level accuracy positioning, which is used to synchronize with image data to determine the spatial location of traffic lights.

[0043] The attitude data measured by the inertial measurement unit includes the vehicle's acceleration, angular velocity, and pitch angle. Specifically, it can be achieved using a combination of gyroscopes and accelerometers to compensate for image jitter caused by vehicle motion and improve the accuracy of multi-source data alignment.

[0044] Specifically, as a vehicle approaches an intersection, a photosensor detects the ambient light intensity in real time and uses this information, combined with a navigation map, to determine the intersection's structural features. For example, if backlighting or shadow areas are detected, the camera's exposure parameters and focal length are automatically adjusted to ensure clear imaging of the traffic light area. The adjusted camera then captures dynamic images at a fixed frame rate, simultaneously recording the vehicle's position and attitude data. Images are then aligned with GPS and attitude data using timestamps; for instance, images captured at the same time are bound to the vehicle's position coordinates, forming a spatiotemporally synchronized continuous image sequence. This provides high-precision spatiotemporally correlated data for subsequent traffic light detection.

[0045] By dynamically adjusting camera parameters and combining multi-source data synchronization, the problem of unstable image quality caused by environmental changes is effectively solved. For example, in strong light environments, overexposure of traffic light areas is avoided by shortening the exposure time. At the same time, during the data alignment stage, attitude data is used to compensate for the impact of vehicle bumps on the image, significantly improving the spatiotemporal consistency of the image sequence.

[0046] Furthermore, prior to data acquisition, the following steps are included: calibrating the camera's internal and external parameters to obtain its precise internal and external parameters; equipping the camera with an automatic aperture adjustment device that can automatically adjust the focal length according to vehicle speed and ambient brightness; and including a temperature sensing module in the camera, which adjusts imaging parameters through a temperature compensation algorithm to reduce the impact of temperature changes on image quality.

[0047] The internal and external parameter calibration refers to determining the camera's internal optical parameters and external mounting position parameters through experimental measurement. Specifically, this can be achieved using a checkerboard calibration method combined with a nonlinear optimization algorithm. This eliminates geometric distortion of the camera lens and establishes a mapping relationship between the image coordinate system and the real-world coordinate system. The automatic aperture adjustment device refers to a mechanical structure capable of dynamically adjusting the optical aperture. Specifically, a closed-loop control algorithm can be used to adjust the focal length in real time based on ambient light intensity sensor data and vehicle speed signals, adapting to the focusing requirements of distant traffic lights at different driving speeds. The temperature compensation algorithm corrects imaging parameters based on real-time temperature data collected by a temperature sensor. Specifically, a thermistor combined with a linear regression model can be used to predict the effect of temperature on the lens refractive index, suppressing image blurring or color shift caused by temperature fluctuations.

[0048] During vehicle startup, the camera undergoes initial calibration using a checkerboard calibration board, establishing a transformation matrix between pixel coordinates and 3D spatial coordinates. While the vehicle is in motion, an automatic aperture adjustment device calculates the optimal focal length parameters based on brightness data collected by a light intensity sensor and vehicle speed signals; for example, increasing the aperture diameter at high speeds to enhance light intake. A temperature sensing module monitors real-time temperature changes within the camera; when the temperature exceeds a set threshold, a compensation algorithm is triggered to adjust white balance parameters and exposure time. Through this collaborative mechanism, stable image quality can be maintained under complex environmental conditions, providing clear image input for subsequent traffic light detection.

[0049] In some specific implementations, the calibration process can use Zhang's calibration method to calculate the radial distortion coefficient of the lens, the automatic adjustment device can integrate a stepper motor to drive the aperture blades, and the temperature compensation algorithm can establish a lens refractive index lookup table at different temperatures. For example, within the operating temperature range of -20℃ to 60℃, a set of compensation parameters is pre-stored every 5℃, and the optimal imaging parameters corresponding to the current temperature are calculated through linear interpolation during real-time operation.

[0050] In another embodiment, the image acquisition device consists of a front-facing camera and a rear-facing camera, which can simultaneously acquire images of traffic lights in both forward and backward directions; the camera mounting bracket has an adjustable tilt angle, which can adjust the camera's pitch angle according to different vehicle speeds to optimize the field of view; a polarizing filter is provided on the camera's optical system to suppress glare caused by rain and backlighting at night, thereby improving the visibility of the traffic lights.

[0051] The front and rear cameras are image acquisition devices installed at the front and rear of the vehicle, respectively. These can be implemented using sensor modules with high dynamic range imaging capabilities. Their function is to cover traffic lights in the direction of vehicle travel and the area behind the vehicle, avoiding missed detections due to blind spots. The adjustable tilt mounting bracket is a mechanical structure with an electric drive, which can be implemented using a servo motor and an angle sensor. By adjusting the camera's tilt angle, the field of view is changed to adapt to the need for capturing traffic lights at different distances at varying vehicle speeds. The polarizing filter is an optical element installed in front of the camera lens, which can be implemented using a linear polarizer or a ring polarizer. It reduces interference from water reflections, glass reflections, and scattering in rain and fog environments by filtering light with specific polarization directions.

[0052] During vehicle operation, the front-facing camera continuously captures images of traffic lights at intersections ahead, while the rear-facing camera simultaneously acquires images of traffic lights at intersections behind, creating an omnidirectional detection capability. When the vehicle accelerates, the mounting bracket drives the camera to adjust its tilt angle downwards to focus on nearby traffic lights; when the vehicle decelerates, the camera tilts upwards to expand the detection range for longer distances. In rainy or nighttime scenarios, a polarizing filter effectively eliminates interference such as reflections from water droplets on the windshield and glare from oncoming headlights, ensuring that the color characteristics of the traffic lights remain clearly discernible in the image. Thus, the dual cameras work together to ensure 360-degree detection coverage, the dynamically adjusted camera angle optimizes target capture capabilities at different vehicle speeds, and polarization processing improves image quality under complex lighting conditions.

[0053] In one embodiment, the step of performing image denoising, illumination equalization, and image enhancement processing on a continuous image sequence to obtain an enhanced image sequence includes converting the color space of each image in the continuous image sequence; applying Gaussian filtering and median filtering to each converted image for denoising; and using an adaptive histogram equalization method to enhance each denoised image to obtain an enhanced image sequence.

[0054] It should be noted that this color space conversion refers to converting the image from RGB color mode to HSV or Lab mode, which are more suitable for feature extraction. Specifically, it can be implemented using the cv2.cvtColor function in the OpenCV library. This conversion can separate the luminance and chrominance components for subsequent processing.

[0055] The Gaussian filtering refers to using a Gaussian kernel to perform a convolution operation on the image to eliminate high-frequency noise. Specifically, a 5×5 kernel with a standard deviation of 1.5 can be used. This step can suppress the interference of sensor noise on subsequent detection.

[0056] The median filtering refers to replacing the current pixel value with the median of the pixel's neighborhood. Specifically, it can be implemented using a 3×3 window. This step can effectively remove salt-and-pepper noise while preserving edge information.

[0057] The adaptive histogram equalization refers to dividing the image into sub-regions and performing histogram equalization independently. Specifically, it can be implemented using the CLAHE algorithm. This step improves the visibility of signal lights in dark areas by enhancing local contrast.

[0058] Specifically, during vehicle movement, the raw images captured by the camera suffer from uneven ambient lighting and motion blur, resulting in loss of detail in the traffic light area. By converting the image to HSV spatial separation of the luminance channel, Gaussian filtering is first applied to eliminate random noise, followed by median filtering to remove isolated noise points. The denoised image is then divided into 8×8 sub-blocks, and each sub-block undergoes independent histogram equalization to enhance local contrast; for example, in backlit scenes, this can recover red traffic light areas obscured by strong light. In the processed image sequence, the edge sharpness and color saturation of the traffic light targets are significantly improved, providing reliable input for subsequent detection algorithms.

[0059] By combining Gaussian filtering and median filtering to achieve multi-level noise reduction, and combining it with local adaptive equalization, it is possible to eliminate complex noise while preserving the shape characteristics of traffic lights, such as effectively suppressing the interference of vehicle headlight glare on the green traffic light area at night.

[0060] In practical applications, the original RGB image is first converted to the HSV (Hue, Saturation, Value) color space. Specifically, this is achieved by calling an image library function (such as OpenCV's cvtColor) and setting the conversion parameter to COLOR_RGB2HSV, converting each consecutive frame of the image from RGB space to HSV space. After conversion, the image's color information (such as the hue values ​​of red, green, and yellow traffic lights) is concentrated in the H channel, while saturation and value are independently represented by the S and V channels, respectively, facilitating subsequent color feature extraction and analysis.

[0061] After color conversion, road regions (ROIs) are cropped first. Specifically, road boundaries are identified based on lane detection algorithms (such as Canny edge detection combined with Hough transform) to determine the road regions corresponding to the vehicle's driving direction. ROIs containing potential traffic light locations (such as the upper 1 / 3 of the image) are cropped out, excluding irrelevant backgrounds such as the sky and buildings, thus reducing the scope of subsequent processing and improving computational efficiency.

[0062] When performing Gaussian filtering on the converted image, Gaussian filtering is used for smoothing. Specifically, the Gaussian kernel size is selected and adjusted according to the image resolution. Then, the Gaussian kernel weights are calculated, and a convolution operation is performed on each frame of HSV image within the ROI to suppress high-frequency noise while preserving the blurriness of the traffic light edges.

[0063] Further, during median filtering denoising, median filtering is used to further denoise the salt-and-pepper noise in the image. Specifically: Set the median filter window size and iterate through each pixel within the ROI; The pixel values ​​within the window are sorted, and the median value is used to replace the center pixel value, effectively eliminating isolated noise points while maintaining the sharpness of the traffic light edges.

[0064] Finally, adaptive histogram equalization is used for enhancement, specifically: The image is divided into multiple non-overlapping sub-blocks, and the histogram of each sub-block is calculated independently. Limit the contrast threshold to prevent local noise from being excessively amplified; Histogram equalization is performed on each sub-block to enhance the contrast of local areas; By merging sub-block boundaries through bilinear interpolation and eliminating inter-block artifacts, an enhanced image sequence with global contrast balance is obtained.

[0065] In one implementation, the step of detecting and locating traffic light targets in the enhanced image sequence based on a high-precision MCD algorithm, and constraining the detection and location results using temporal correlation constraints between consecutive frames to obtain traffic light candidate regions, includes: dividing each image in the enhanced image sequence into multiple grid regions, and selecting multiple candidate enhanced images of potential traffic light candidate regions from the multiple grid regions based on the basic characteristics of traffic lights; performing correlation analysis and confidence calculation on the pixel features of each candidate enhanced image using the high-precision MCD algorithm, and determining the target enhanced image based on the analysis results and confidence scores, wherein the correlation analysis includes matching the color histogram and shape template of the traffic lights; clustering and fusion processing of the grid regions in the multiple target enhanced images using coherence constraints, and predicting and updating the light spot trajectories in the grid regions using tracking and Kalman filtering methods to output traffic light candidate regions.

[0066] The grid region division refers to dividing the input image into multiple equally spaced or non-uniformly distributed rectangular sub-regions. Specifically, it can be implemented using a sliding window or region growing algorithm, thereby improving the efficiency of local feature analysis by reducing the detection range.

[0067] The correlation analysis refers to calculating the similarity index between pixel features and preset traffic light templates. Specifically, it can be implemented using histogram cross kernel function or convolution operation to quantify the degree of matching between candidate regions and target features.

[0068] The coherence constraint refers to using temporal continuity to verify the spatial consistency of candidate regions. Specifically, it can be achieved by optical flow tracking or motion trajectory prediction, thereby improving detection stability by eliminating instantaneous interference.

[0069] The Kalman filtering method refers to dynamically estimating the target position based on the state equation and the observation equation. Specifically, it can be implemented using a linear motion model or an acceleration model, and improves the accuracy of trajectory prediction by fusing information from multiple frames.

[0070] Specifically, the enhanced image sequence is first segmented into grid regions, and candidate regions that may contain traffic lights are selected based on preset size thresholds and positional rules. Then, color histogram matching and shape template comparison are performed within each candidate region. By calculating the similarity score between the pixel distribution and the standard traffic light model, candidate regions with confidence scores higher than a threshold are selected. Further, spatial clustering is performed on the candidate regions detected in multiple consecutive frames to eliminate isolated noise points and merge adjacent regions. Finally, a Kalman filter is used to model the motion trajectory of the candidate regions, and combined with historical position data, the possible position of the traffic light in the current frame is predicted, thereby correcting the offset error in the detection results.

[0071] It should be noted that the image grid division and potential region selection specifically involves uniformly dividing each frame of the enhanced image into multiple equally sized grid regions. Based on the basic characteristics of traffic lights, grid regions that meet the characteristics are initially selected as potential traffic light candidate regions. For example, red or green regions with a saturation of ≥70% and a brightness of ≥50% in the HSV color space, or grids with an aspect ratio close to 1:1 or 1:2, are preferentially selected to generate candidate enhanced images.

[0072] For pixel feature analysis and confidence calculation, specifically, for each candidate enhanced image, a lightweight convolutional neural network (CNN) deployed in the vehicle edge computing unit extracts pixel-level features (such as color distribution, edge gradient, and texture information); using a high-precision MCD algorithm, combined with the color histogram template of traffic lights (pre-stored HSV distribution range of typical traffic lights) and shape template (pre-stored contour features of light shapes such as circles and arrows), correlation matching is performed. Color matching: Calculate the color histogram similarity (e.g., Bach distance) between the candidate region and the template. If the similarity is ≥0.8, the color is marked as reliable. Shape matching: Extract features such as contour closure (≥0.9) and roundness (≥0.85 for circular lights, ≥0.75 for arrow lights) of candidate regions and compare them with the template; Confidence calculation: Combining the color and shape matching results with the confidence score (0-1) of the candidate region output by the MCD algorithm training model, regions with a confidence score ≥0.7 are selected as the target enhancement image.

[0073] After completing the confidence calculation above, multi-frame temporal correlation constraint processing is performed, which involves constraining the target enhancement image using the temporal correlation between consecutive frames. Specifically, this includes: Optical flow tracking prediction: For the traffic light target detected in the previous frame, use optical flow methods (such as the Lucas-Kanade algorithm) to calculate its displacement vector in the current frame, predict its position, and narrow down the search range of the current frame; Background difference detection of new regions: For regions not covered by optical flow tracking, moving targets (such as newly added or temporarily adjusted traffic lights) are identified by the background difference algorithm (subtracting the current frame from the historical average background frame) to extract dynamic candidate regions; Kalman filter trajectory smoothing: A state model is established for the position coordinates (x,y) of the same candidate region in multiple frames. The motion trajectory is predicted by Kalman filter to smooth the position fluctuations caused by vehicle shaking or camera shaking, thereby improving positioning stability. Non-maximum suppression (NMS): For detection boxes that overlap in multiple frames (IoU ≥ 0.5), retain the region with the highest confidence and eliminate redundant detections.

[0074] Finally, candidate region clustering and fusion output are performed. Specifically, based on multi-frame detection results, candidate regions with consistent color and shape features are clustered (e.g., using the DBSCAN algorithm), and spatially adjacent regions with similar features are merged. The feature weights are dynamically adjusted using an online learning mechanism (e.g., increasing the weight of brightness features in nighttime scenes and increasing the weight of edge features in rainy scenes), and an attention mechanism is introduced to strengthen features strongly correlated with traffic lights (e.g., high-saturation red regions). Finally, a high-confidence (≥0.85) and coherent traffic light candidate region is output as the input for subsequent state determination.

[0075] By using grid partitioning and confidence filtering mechanisms, irrelevant background areas can be quickly eliminated; by using temporal correlation constraints and trajectory prediction, false positives and false negatives caused by vehicle bumps or target occlusion can be eliminated; by using multi-scale feature fusion and dynamic model updates, the system can adapt to changes in traffic light morphology at different distances and viewpoints, ultimately achieving highly robust dynamic traffic light detection.

[0076] In one implementation, the step of using a high-precision MCD algorithm to perform correlation analysis and confidence calculation on the pixel features of each candidate enhanced image, and determining the target enhanced image based on the analysis results and confidence scores, includes: using a lightweight convolutional neural network model deployed in the vehicle-side edge computing unit to extract the pixel features of each candidate enhanced image; using the high-precision MCD algorithm to calculate the confidence score of the pixel features of each candidate enhanced image as traffic lights; matching the extracted pixel features with the color histogram and shape template of traffic lights, and identifying the target enhanced image that conforms to the morphological features of traffic lights based on the matching results and confidence scores.

[0077] In this embodiment, the lightweight convolutional neural network model refers to a neural network architecture that has undergone model compression and optimization. Specifically, it can be implemented using depthwise separable convolution or channel pruning techniques, reducing the number of parameters to 1 / 5 to 1 / 3 of the standard model. Low-latency feature extraction can be achieved in edge computing units. This model extracts the texture, edge, and color distribution features of candidate regions through multi-layer convolution operations for subsequent confidence calculation.

[0078] The high-precision MCD algorithm refers to a detection algorithm based on multi-dimensional feature fusion. Specifically, it calculates the similarity between the color histogram of the candidate region and a preset traffic light template, and combines this with the shape contour matching degree to generate a comprehensive confidence score. This algorithm reduces the probability of color misjudgment under complex lighting conditions by introducing color stability constraints and shape space distribution constraints.

[0079] Color histogram and shape template matching involves mapping candidate region pixels to the HSV space, statistically analyzing their saturation and brightness distribution, and comparing it with a preset traffic light color threshold range. Simultaneously, it verifies whether the candidate region's outline conforms to the geometric features of a traffic light by calculating the similarity between the candidate region's outline and standard circular and rectangular templates. This matching process uses a dual feature verification mechanism to eliminate interference from unstructured targets.

[0080] Specifically, as a vehicle approaches an intersection, the edge computing unit processes the candidate augmented image in real time. First, a lightweight convolutional neural network extracts the depth features of the candidate region, generating a feature vector containing color distribution, edge intensity, and local contrast. Then, a high-precision MCD algorithm inputs the feature vector into a multilayer perceptron to calculate the probability that it belongs to a traffic light and outputs a confidence score. Simultaneously, the algorithm analyzes the overlap between the HSV color histogram of the candidate region and preset red and green threshold ranges, and uses a shape template matching algorithm to verify whether its contour conforms to the standard traffic light shape. When the confidence score exceeds the threshold and both color and shape matching meet the requirements, the candidate region is determined to be the target augmented image and enters the subsequent state recognition process.

[0081] By employing a lightweight model to automate feature extraction and combining it with a high-precision MCD algorithm for dual verification of color and shape features, this approach effectively addresses the issue of decreased recognition rates caused by color distortion and shape variations in traffic lights in dynamic environments. For instance, in backlit scenarios, traditional methods may cause color threshold failure due to sudden brightness changes, while this solution maintains stable recognition performance through robust features extracted via a convolutional neural network.

[0082] In one implementation, the step of extracting color and shape features from the traffic light candidate regions, determining the traffic light state based on the extracted features, and outputting the traffic light detection results includes: inputting each traffic light candidate region into an edge computing unit, performing HSV color space conversion and edge detection, and extracting the saturation, brightness, and shape contour of each traffic light candidate region; applying predefined color discrimination rules to the extracted saturation and brightness, and performing direction template matching on the shape contour to determine the traffic light color state and direction state; writing the traffic light color state and direction state of each traffic light candidate region into the traffic light state cache queue, and using a sliding time window mechanism to fuse the states of multiple consecutive frames, and outputting the traffic light detection results.

[0083] The HSV color space conversion refers to converting an image from the RGB color model to a model in which hue, saturation, and brightness are represented independently. This can be achieved using a color space conversion algorithm. This method can effectively separate color information from light intensity, making it easier to perform color discrimination based on saturation in subsequent processes.

[0084] Edge detection refers to identifying object boundaries by calculating the gradient of image grayscale changes. For example, the Canny operator is used to extract shape contours. This step can accurately capture the geometric structural features of traffic lights. Predefined color discrimination rules refer to classification conditions established based on the standard color range of traffic lights. For example, setting saturation threshold ranges for red and green, and filtering target color regions by comparing thresholds.

[0085] Directional template matching refers to calculating the similarity between the extracted contour shape and preset arrow or circular templates. For example, a normalized cross-correlation algorithm can be used to evaluate the degree of matching between the contour and the template, thereby determining the directional status of the traffic light.

[0086] The sliding time window mechanism refers to weighted fusion of state data from multiple consecutive frames. For example, time series averaging can be used to eliminate misjudgments in a single frame. This mechanism can improve the stability of state determination.

[0087] Specifically, after the traffic light candidate regions are input into the edge computing unit, they first undergo HSV color space conversion, decomposing the image into three independent channels: hue, saturation, and brightness. In the saturation channel, candidate regions are filtered by setting red-green color threshold ranges, while overexposed or underexposed interference regions are eliminated in the brightness channel. The edge detection algorithm extracts the contours of the candidate regions, obtaining closed geometric shape features. The direction template matching module compares the extracted contours with preset arrow direction templates, calculating a shape similarity score. The color discrimination rules and direction matching results are written to a cache queue. A sliding time window performs a weighted average of the state data from the most recent five frames; when the color state of the same traffic light is consistent in three consecutive frames, it is considered a valid detection result.

[0088] In one implementation, predefined color discrimination rules are applied to the extracted saturation and brightness, and directional template matching is performed on the shape contour to determine the color and directional state of the traffic light. This includes: using a color stability filter based on statistical fluctuation range to determine whether the saturation and brightness of each traffic light candidate region are within a reliable fluctuation range; if not, selecting target traffic light candidate regions with similar colors based on saturation and brightness, and performing closure calculation and distribution symmetry analysis on the shape contour of the target traffic light candidate regions; and performing directional template matching based on the closed shape contour to determine the color and directional state of the traffic light.

[0089] Among them, the color stability filter refers to establishing the fluctuation range threshold of color parameters through statistical methods. Specifically, the confidence interval can be calculated by using the mean and standard deviation of saturation and brightness within a sliding window, which is used to determine whether the color parameters of the current candidate region are stable.

[0090] Closure calculation refers to the quantitative evaluation of the continuity of a shape profile. Specifically, it can be determined by calculating the ratio of the perimeter of the profile to its area to determine whether the profile forms a closed region, thereby eliminating the interference of noise or broken edges.

[0091] Distribution symmetry analysis refers to the symmetry evaluation of the spatial distribution characteristics of a shape profile. Specifically, it can be achieved by using mirror symmetry axis detection or centroid offset calculation to verify whether the profile conforms to the standard geometric shape of a traffic light.

[0092] Directional template matching refers to comparing the similarity between a closed contour and a predefined arrow, circle, or rectangle directional template. Specifically, normalized cross-correlation algorithms or feature point matching methods can be used to determine the directional state of a traffic light.

[0093] Specifically, in the color discrimination process for traffic light candidate areas, a color stability filter is first used to screen candidate areas whose saturation and brightness are within a reasonable fluctuation range. If the color parameters of a candidate area exceed a threshold, areas with similar colors are further screened to avoid misjudgments caused by sudden changes in illumination or glare interference. For the screened target areas, non-closed contours, such as irregular shapes like tree branches or vehicle lights, are eliminated through closure calculations. Subsequently, the distribution symmetry of closed contours is analyzed to verify whether they conform to the standard symmetrical structure of traffic lights, such as circular or regularly arranged light groups. Finally, contours that meet the symmetry requirements are matched with pre-stored direction templates to determine the directional state of the traffic light, such as a left-turn arrow or a straight-ahead indicator.

[0094] In summary, this application's embodiments, through real-time acquisition of continuous image sequences using a high-resolution camera, can capture the complete process of dynamic changes in traffic lights, providing a continuous temporal data foundation for subsequent analysis. Secondly, image denoising, illumination equalization, and enhancement processing effectively eliminate the influence of environmental noise and illumination deviations, improving the distinction between the traffic light target and the background, and providing high-quality input data for the MCD algorithm. Furthermore, based on a high-precision MCD algorithm combined with continuous frame temporal correlation constraints, it can filter out single-frame false detections through multi-frame information fusion and capture the continuity of traffic light movement, reducing the risk of missed detections. Finally, through the extraction of color and shape features and multi-frame state fusion judgment, accurate identification of the color and direction status of traffic lights is achieved, avoiding misjudgments caused by single-frame flickering or viewpoint deviations. In conclusion, the synergistic effect of the above technical features improves the detection accuracy of the method at complex intersections compared to traditional MCD algorithms, effectively solving the problems of detection lag, missed detections, and map update delays in existing technologies.

[0095] For the corresponding method embodiments described above, see [link to relevant documentation]. Figure 2 The diagram shown illustrates a dynamic traffic light detection device based on a high-precision MCD algorithm. The device includes: The acquisition module 210 is used to acquire a continuous sequence of images in a traffic environment in real time using a high-resolution camera installed on the vehicle, wherein the continuous sequence of images includes dynamically changing traffic light targets. Enhancement module 220 is used to perform image denoising, illumination equalization and image enhancement processing on the continuous image sequence to obtain an enhanced image sequence; The detection module 230 is used to detect and locate traffic light targets in the enhanced image sequence based on the high-precision MCD algorithm, and to constrain the detection and location results by using the temporal correlation constraints between consecutive frames to obtain traffic light candidate regions. The determination module 240 is used to extract color and shape features from the candidate area of ​​the traffic signal light, determine the state of the traffic signal light based on the extracted features, and output the detection result of the traffic signal light. The detection result includes determining the color state and direction state of each traffic signal light.

[0096] Optionally, the aforementioned acquisition module 210 is specifically used for: When the vehicle approaches an intersection, the environmental parameters of the vehicle's location are analyzed, and the operating parameters of the camera mounted on the front of the vehicle are adjusted based on the environmental parameters. The operating parameters include at least one of exposure parameters and distance. The system uses an adjusted camera to capture continuous dynamic images of the intersection in real time, as well as record the GPS positioning information of the vehicles and the attitude data measured by the inertial measurement unit. The GPS positioning information, attitude data and continuous dynamic images are then aligned with the timestamps to obtain a continuous image sequence.

[0097] Optionally, the aforementioned enhancement module 220 is specifically used for: Perform color space conversion on each image in the continuous image sequence; The converted images were then de-noiseed by applying Gaussian filtering followed by median filtering. An enhanced image sequence is obtained by using an adaptive histogram equalization method to enhance each of the denoised images.

[0098] Optionally, the detection module 230 described above is specifically used for: Each image in the enhanced image sequence is divided into multiple grid regions, and candidate enhanced images of multiple potential traffic light candidate regions are selected from the multiple grid regions based on the basic characteristics of traffic lights; The high-precision MCD algorithm is used to perform correlation analysis and confidence calculation on the pixel features of each candidate enhanced image, and the target enhanced image is determined based on the analysis results and confidence. The correlation analysis includes matching the color histogram and shape template of the traffic lights. Clustering and fusion processing of grid regions in multiple target augmentation images is performed using coherence constraints, and the trajectory of light points in the grid regions is predicted and updated using tracking and Kalman filtering methods to output candidate regions for traffic lights.

[0099] Optionally, the detection module 230 described above is specifically used for: A lightweight convolutional neural network model deployed in the vehicle-side edge computing unit is used to extract pixel features for each candidate enhanced image; Using a high-precision MCD algorithm, the confidence level of the pixel features of each candidate enhanced image as a traffic light is calculated; The extracted pixel features are matched with the color histogram and shape template of traffic lights, and the target enhanced image that conforms to the morphological features of traffic lights is identified based on the matching results and confidence scores.

[0100] Optionally, the aforementioned determination module 240 is specifically used for: Each traffic light candidate region is input into the edge computing unit, where HSV color space conversion and edge detection are performed to extract the saturation, brightness, and shape contour of each traffic light candidate region. Predefined color discrimination rules are applied to the extracted saturation and brightness, and directional template matching is performed on the shape contour to determine the color and directional state of the traffic light. The color and direction states of each traffic light candidate area are written into the state buffer queue of the traffic light, and the states of multiple consecutive frames are fused using a sliding time window mechanism to output the detection results of the traffic lights.

[0101] Optionally, the aforementioned determination module 240 is specifically used for: A color stability filter based on statistical fluctuation range is used to determine whether the saturation and brightness of each traffic light candidate area are within a reliable fluctuation range. If not, then based on saturation and brightness, select target traffic light candidate areas with similar colors, and perform closure calculation and distribution symmetry analysis on the shape contour of the target traffic light candidate areas; Based on the closed shape contour, perform directional template matching to determine the color and direction status of the traffic light.

[0102] The device provided in the above embodiments acquires continuous image sequences in real time using an onboard high-resolution camera; performs denoising, illumination equalization, and enhancement processing on the image sequences; performs target detection and localization based on a high-precision MCD algorithm, and filters candidate regions by combining continuous frame temporal correlation constraints; extracts color and shape features of the candidate regions for state determination, and outputs the color and direction status of the traffic lights. This effectively solves the problem of low traffic light recognition accuracy in complex dynamic environments, and has significant advantages in improving detection accuracy, enhancing environmental adaptability, and ensuring the safety of autonomous driving.

[0103] This embodiment also provides an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor. The processor executes the machine-executable instructions to implement the aforementioned dynamic traffic light detection method based on a high-precision MCD algorithm. This electronic device can be a server or a terminal device.

[0104] See Figure 3As shown, the electronic device includes a processor 100 and a memory 101. The memory 101 stores machine-executable instructions that can be executed by the processor 100. The processor 100 executes the machine-executable instructions to implement the above-mentioned dynamic traffic light detection method based on the high-precision MCD algorithm.

[0105] Furthermore, Figure 3 The electronic device shown also includes a bus 102 and a communication interface 103, with the processor 100, the communication interface 103 and the memory 101 connected via the bus 102.

[0106] The memory 101 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 102 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0107] The processor 100 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 100 or by instructions in software form. The processor 100 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 101. The processor 100 reads the information in memory 101 and, in conjunction with its hardware, completes the steps of the dynamic traffic light detection method based on the high-precision MCD algorithm described in the aforementioned embodiment.

[0108] This embodiment also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions cause the processor to implement the above-described dynamic traffic light detection method based on a high-precision MCD algorithm.

[0109] The computer program product of the dynamic traffic light detection method, apparatus, device and storage medium provided in the embodiments of this disclosure includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0110] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0111] Furthermore, in the description of the embodiments of this disclosure, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this disclosure based on the specific circumstances.

[0112] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] In the description of this disclosure, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this disclosure and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this disclosure. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0114] Finally, it should be noted that the above embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A dynamic traffic light detection method based on a high-precision MCD algorithm, characterized in that, The method includes: A continuous sequence of images of the traffic environment is captured in real time by a high-resolution camera mounted on the vehicle, wherein the continuous sequence of images includes dynamically changing traffic light targets; The continuous image sequence is subjected to image denoising, illumination equalization and image enhancement processing to obtain an enhanced image sequence; The enhanced image sequence is used to detect and locate traffic light targets based on the high-precision MCD algorithm, and the detection and localization results are constrained by the temporal correlation between consecutive frames to obtain traffic light candidate regions. The candidate regions for traffic lights are subjected to color and shape feature extraction. Based on the extracted features, the status of the traffic lights is determined, and the detection results of the traffic lights are output. The detection results include determining the color status and direction status of each traffic light.

2. The dynamic traffic light detection method according to claim 1, characterized in that, The method of acquiring a continuous sequence of images of the traffic environment in real time using a high-resolution camera installed on the vehicle includes: When the vehicle approaches an intersection, the environmental parameters of the vehicle's location are analyzed, and the operating parameters of the camera mounted on the front of the vehicle are adjusted based on the environmental parameters. The operating parameters include at least one of exposure parameters and distance. The system uses an adjusted camera to capture continuous dynamic images of the intersection in real time, as well as record the GPS positioning information of the vehicles and the attitude data measured by the inertial measurement unit. The GPS positioning information, attitude data and continuous dynamic images are then aligned with the timestamps to obtain a continuous image sequence.

3. The dynamic traffic light detection method according to claim 1, characterized in that, The step of performing image denoising, illumination equalization, and image enhancement processing on the continuous image sequence to obtain an enhanced image sequence includes: Perform color space conversion on each image in the continuous image sequence; The converted images were then de-noiseed by applying Gaussian filtering followed by median filtering. An enhanced image sequence is obtained by using an adaptive histogram equalization method to enhance each of the denoised images.

4. The dynamic traffic light detection method according to claim 1, characterized in that, The enhanced image sequence is processed using a high-precision MCD algorithm to detect and locate traffic light targets. The detection and localization results are constrained by temporal correlation between consecutive frames to obtain candidate regions for traffic lights, including: Each image in the enhanced image sequence is divided into multiple grid regions, and candidate enhanced images of multiple potential traffic light candidate regions are selected from the multiple grid regions based on the basic characteristics of traffic lights; The high-precision MCD algorithm is used to perform correlation analysis and confidence calculation on the pixel features of each candidate enhanced image, and the target enhanced image is determined based on the analysis results and confidence. The correlation analysis includes matching the color histogram and shape template of the traffic lights. Clustering and fusion processing of grid regions in multiple target augmentation images is performed using coherence constraints, and the trajectory of light points in the grid regions is predicted and updated using tracking and Kalman filtering methods to output candidate regions for traffic lights.

5. The dynamic traffic light detection method according to claim 4, characterized in that, The process of using a high-precision MCD algorithm to perform correlation analysis and confidence calculation on the pixel features of each candidate enhanced image, and determining the target enhanced image based on the analysis results and confidence scores, includes: A lightweight convolutional neural network model deployed in the vehicle-side edge computing unit is used to extract pixel features for each candidate enhanced image; Using a high-precision MCD algorithm, the confidence level of the pixel features of each candidate enhanced image as a traffic light is calculated; The extracted pixel features are matched with the color histogram and shape template of traffic lights, and the target enhanced image that conforms to the morphological features of traffic lights is identified based on the matching results and confidence scores.

6. The dynamic traffic light detection method according to claim 1, characterized in that, The process of extracting color and shape features from the candidate regions of the traffic lights, determining the state of the traffic lights based on the extracted features, and outputting the detection results of the traffic lights includes: Each traffic light candidate region is input into the edge computing unit, where HSV color space conversion and edge detection are performed to extract the saturation, brightness, and shape contour of each traffic light candidate region. Predefined color discrimination rules are applied to the extracted saturation and brightness, and directional template matching is performed on the shape contour to determine the color and directional state of the traffic light. The color and direction states of each traffic light candidate area are written into the state buffer queue of the traffic light, and the states of multiple consecutive frames are fused using a sliding time window mechanism to output the detection results of the traffic lights.

7. The dynamic traffic light detection method according to claim 6, characterized in that, The process of applying predefined color discrimination rules to the extracted saturation and brightness, and performing direction template matching on the shape contour to determine the color and direction states of the traffic light includes: A color stability filter based on statistical fluctuation range is used to determine whether the saturation and brightness of each traffic light candidate area are within a reliable fluctuation range. If not, then based on saturation and brightness, select target traffic light candidate areas with similar colors, and perform closure calculation and distribution symmetry analysis on the shape contour of the target traffic light candidate areas; Based on the closed shape contour, perform directional template matching to determine the color and direction status of the traffic light.

8. A dynamic traffic light detection device based on a high-precision MCD algorithm, characterized in that, The device includes: The acquisition module is used to acquire a continuous sequence of images in the traffic environment in real time using a high-resolution camera installed on the vehicle, wherein the continuous sequence of images includes dynamically changing traffic light targets; The enhancement module is used to perform image denoising, illumination equalization and image enhancement processing on the continuous image sequence to obtain an enhanced image sequence; The detection module is used to detect and locate traffic light targets in the enhanced image sequence based on the high-precision MCD algorithm, and to constrain the detection and localization results by utilizing the temporal correlation constraints between consecutive frames to obtain traffic light candidate regions. The determination module is used to extract color and shape features from the candidate areas of the traffic lights, determine the state of the traffic lights based on the extracted features, and output the detection results of the traffic lights. The detection results include determining the color state and direction state of each traffic light.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the dynamic traffic light detection method based on the high-precision MCD algorithm as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the dynamic traffic light detection method based on the high-precision MCD algorithm as described in any one of claims 1-7.

Images