Ultra-wide-angle pedestrian detection method and system based on visual angle fusion and intelligent focusing

Abstract

The invention discloses an ultra-wide-angle pedestrian detection method and system based on visual angle fusion and intelligent focusing, and belongs to the field of computer vision and target detection, and the method comprises the steps: obtaining to-be-detected ultra-wide-angle pedestrian data, and carrying out the preprocessing and blind area visual angle fusion; inputting the fused data into the trained pedestrian detection model to obtain a pedestrian detection result; wherein the training of the pedestrian detection model comprises the following steps: constructing an ultra-wide-angle pedestrian data set; a pedestrian detection model is built, the pedestrian detection model comprises an intelligent focusing module, an intelligent blind area pedestrian detection module and a driver warning module, the built pedestrian detection model is trained through the built ultra-wide-angle pedestrian data set, and the trained pedestrian detection model is obtained. According to the invention, the limitation of a traditional pedestrian detection system in a wide-angle view field and a high-dynamic environment is broken through, the pedestrian sensing capability is remarkably improved, traffic accidents caused by a ghost probe are effectively reduced, and the system has a wide application prospect in the fields of automatic driving, intelligent security and protection and the like.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and target detection, and specifically relates to an ultra-wide-angle pedestrian detection method and system based on perspective fusion and intelligent focusing. Background Art

[0002] With the rapid development of intelligent transportation systems and autonomous driving technologies, pedestrian detection has become a core component of road safety. However, existing technologies still have significant limitations in complex and dynamic scenarios. In particular, traditional solutions struggle to meet the requirements for high robustness, wide coverage, and real-time response when responding to sudden and dangerous incidents.

[0003] First, existing pedestrian detection systems and dashcams rely on single-view cameras with a limited field of view, making it difficult to fully cover blind spots around the vehicle (such as beams, columns, and the sides and rear of the vehicle). While the introduction of ultra-wide-angle cameras can address this limited field of view and expand the detection range, image distortion is particularly prominent in the edge regions, limiting detection accuracy in ultra-wide-angle scenarios, especially at the edges. Although some algorithms have incorporated wide-angle distortion correction techniques, image quality and object detection performance remain significantly deficient in ultra-wide-angle scenarios, failing to effectively overcome the field of view limitations. Second, during turns, pedestrians in the same-side blind spot can be easily overlooked due to the inner wheel difference and the driver's limited field of view. Existing detection systems typically use a static target weighting strategy, failing to dynamically adjust the detection priority on both sides of the vehicle based on the vehicle's real-time motion state, namely the centripetal acceleration during the turn. During a turn, the system cannot proactively focus on potentially dangerous areas within the inner wheel difference coverage area, resulting in a delayed response to sudden pedestrians and failing to meet the required time window for emergency braking in critical situations. In summary, due to problems such as limited viewing angle, insufficient distortion correction, and poor adaptability to dynamic scenes, existing technologies are unable to effectively deal with high-risk scenarios such as "ghosting" in complex traffic environments. There is an urgent need for an innovative solution that takes into account wide-angle coverage, high-precision correction, and intelligent dynamic perception.

[0004] Therefore, it is necessary to design an ultra-wide-angle pedestrian detection method and system based on perspective fusion and intelligent focusing to address the above problems. Summary of the Invention

[0005] The purpose of the present invention is to address the limitations of traditional pedestrian detection systems in wide-angle fields of view and high-dynamic environments, and to provide an ultra-wide-angle pedestrian detection method and system based on perspective fusion and intelligent focusing. By drawing on the bionic structure of nature and adopting an optimization strategy of multi-perspective fusion and distortion correction, the perspective limitations of traditional visual detection systems are broken through, and the detection accuracy of ultra-wide-angle fields of view is improved. Through the preprocessing module, blind spot perspective fusion module, intelligent focusing module, intelligent blind spot pedestrian detection module, and driver warning module, the robustness of target perception and detection is significantly improved, and it has broad application prospects in the fields of autonomous driving, intelligent security, etc.

[0006] According to one aspect of this specification, a method for ultra-wide-angle pedestrian detection based on perspective fusion and intelligent focusing is provided, including:

[0007] Obtain ultra-wide-angle pedestrian data to be tested, perform pre-processing and blind spot perspective fusion;

[0008] The fused data is input into a trained pedestrian detection model to obtain a pedestrian detection result; wherein the training of the pedestrian detection model includes:

[0009] Build an ultra-wide-angle pedestrian dataset;

[0010] Build a pedestrian detection model, including: an intelligent focusing module, which adds visual blind spot markers to input data and dynamically adjusts the detection weights of the left and right blind spots using acceleration sensors; an intelligent blind spot pedestrian detection module, which optimizes the calculation path through an intelligent blind spot matching strategy and adjusts the detection weights of the left and right blind spots based on left and right acceleration information during real-time detection; and a driver warning module, which alerts the driver based on the results of the intelligent blind spot pedestrian detection.

[0011] The constructed ultra-wide-angle pedestrian dataset is used to train the constructed pedestrian detection model to obtain a trained pedestrian detection model.

[0012] Furthermore, pre-processing and blind spot perspective fusion are performed, including:

[0013] Preprocess the ultra-wide-angle pedestrian data, including denoising, normalization, and resizing;

[0014] The pre-processed data is input into the blind spot view fusion module for splicing and fusion to obtain a large-view video stream.

[0015] Furthermore, the intelligent focusing module further includes:

[0016] Attention recognition module, used to analyze image texture and visual blind spot characteristics, and locate the visual blind spots that require priority attention;

[0017] Visual blind spot calculation module, used to automatically enhance attention to targets in visual blind spots based on real-time scene complexity;

[0018] The auxiliary acceleration sensor module is used to capture the acceleration on the left and right sides of the vehicle based on the vehicle's dynamic behavior and add enhanced identification to the left and right blind spot video streams.

[0019] Furthermore, the intelligent blind spot pedestrian detection module further includes:

[0020] The blind spot weight assessment module is used to combine the driver's position information to analyze and output the weight ratio of the left and right blind spots in real time;

[0021] The dynamic resource allocation module is used to intelligently adjust internal computing resources based on the weight ratio of the left and right blind spots and reasonably allocate limited computing power.

[0022] Furthermore, the identification weight formula of the visual blind spot calculation module is:

[0023]

[0024] in, is the image point The visual blind spot identification weight, is the visual attention response computed by the network.

[0025] Furthermore, the blind spot perspective fusion module further includes:

[0026] Frame stitching parameter calculation module, used to extract feature points and calculate the optimal stitching line;

[0027] The video frame fusion module is used to quickly stitch together a large-viewing angle video stream based on the optimal stitching line.

[0028] According to one aspect of this specification, an ultra-wide-angle pedestrian detection system based on perspective fusion and intelligent focusing is provided, comprising:

[0029] The data acquisition module is used to obtain the ultra-wide-angle pedestrian data to be tested, perform pre-processing and blind spot perspective fusion;

[0030] The pedestrian detection module is used to input the fused data into the trained pedestrian detection model to obtain pedestrian detection results. The training of the pedestrian detection model includes:

[0031] Build an ultra-wide-angle pedestrian dataset;

[0032] Build a pedestrian detection model, including: an intelligent focusing module, which adds visual blind spot markers to input data and dynamically adjusts the detection weights of the left and right blind spots using acceleration sensors; an intelligent blind spot pedestrian detection module, which optimizes the calculation path through an intelligent blind spot matching strategy and adjusts the detection weights of the left and right blind spots based on left and right acceleration information during real-time detection; and a driver warning module, which alerts the driver based on the results of the intelligent blind spot pedestrian detection.

[0033] The constructed ultra-wide-angle pedestrian dataset is used to train the constructed pedestrian detection model to obtain a trained pedestrian detection model.

[0034] According to one aspect of this specification, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing are implemented.

[0035] According to one aspect of the present specification, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing are implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. To address issues such as limited viewing angle, insufficient distortion correction, and poor adaptability to dynamic scenes, the present invention overcomes the limitations of traditional pedestrian detection systems in wide-angle viewing angles and high-dynamic environments through a preprocessing module, a blind spot viewing angle fusion module, an intelligent focusing module, an intelligent blind spot pedestrian detection module, and a driver warning module. This significantly improves the robustness of target perception and detection, and has broad application prospects in the fields of autonomous driving, intelligent security, and so on.

[0038] 2. By drawing on the bionic structures of nature and adopting an optimization strategy of multi-perspective fusion and distortion correction, this invention breaks through the viewing angle limitations of traditional visual inspection systems and improves the detection accuracy of ultra-wide-angle fields of view. In addition, a distortion correction algorithm is designed specifically for vehicle blind spots to ensure stable detection under different lighting conditions and complex scenarios.

[0039] 3. The present invention uses an intelligent focusing module, based on the optimization of human eye visual characteristics and the intelligent focusing mechanism of left and right acceleration sensors. By dynamically adjusting the target detection weight and combining the data input from the left and right acceleration sensors, it can perceive the vehicle's motion status in real time and give priority to the characteristics of pedestrians in potentially dangerous areas under turning conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0042] Figure 2 This is a flow chart of blind spot view fusion according to an embodiment of the present invention;

[0043] Figure 3 This is a structural diagram of the intelligent focusing module according to an embodiment of the present invention;

[0044] Figure 4 This is a structural diagram of an intelligent blind spot pedestrian detection module according to an embodiment of the present invention;

[0045] Figure 5 This is a structural diagram of a driver warning module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] like Figure 1 As shown, an embodiment of the present invention provides an ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing, including: step S1, collecting video streams required for pedestrian detection and preprocessing; step S2, step S2: blind spot perspective video stream splicing and fusion, the preprocessed blind spot video stream input row image splicing and alignment to obtain a wide-angle image; step S3, intelligent focusing, by simulating the visual characteristics of the human eye, adding visual blind spot marks to the image outside the human eye's visual range, and dynamically adjusting the detection weights of the left and right blind spots with the assistance of the acceleration sensor; step S4, intelligent blind spot pedestrian detection, through the intelligent blind spot matching strategy, optimizing the calculation path of the deep learning network model, and adjusting the detection weights of the left and right blind spots according to the left and right acceleration information during the real-time detection process; step S5, driver warning module, according to the results of the intelligent blind spot pedestrian detection, by controlling the instrument panel warning and audio warning to remind the driver that a "ghost peeking" situation has occurred, and the driver needs to brake immediately and pay attention to the surrounding situation.

[0048] Specifically, preprocessing includes denoising, normalization, and image resizing to ensure the consistency and accuracy of the video stream input data. These preprocessing operations ensure the accuracy of subsequent image stitching and object detection.

[0049] Specifically, the embodiment of the present invention also provides a blind area view stream splicing fusion, including a frame splicing parameter calculation module and a video frame fusion module. Figure 2 As shown, the frame splicing parameter calculation module includes a feature extraction unit and a stitching line calculation unit. The feature extraction unit takes as input the image information of the first frame of the blind spot video stream and outputs the extracted feature points. The stitching line calculation unit takes as input the feature points of the first frame of the blind spot video stream and outputs the optimal stitching line for the wide-angle image formed by matching the feature points of multiple images. The video frame fusion module takes as input the optimal stitching line calculated by the frame splicing parameter calculation module and outputs the subsequent wide-angle video stream quickly spliced ​​according to the stitching line.

[0050] Specifically, in step S2, the frame stitching parameter calculation module extracts feature points from the first frame image information of multiple video streams based on the feature extraction unit to obtain image feature points, and then calculates the optimal stitching line of the multiple images through the stitching line calculation unit, and finally outputs a wide-angle image.

[0051] Specifically, the feature extraction unit uses the Speeded Up Robust Features (SURF) algorithm to extract feature points. The SURF algorithm detects the extreme points of the image through the Hessian matrix. The binary function The Hessian where is is shown in formula (1).

[0052] (1)

[0053] in, The image is at the pixel point The pixel value at .

[0054] Specifically, the Hessian matrix discriminant is shown in formula (2).

[0055] (2)

[0056] Where det(H) is the determinant of the Hessian matrix. When det(H) is greater than 0, it is the extreme point. The characteristic of the Hessian matrix expression of this pixel in the Gaussian pyramid is shown in Equation (3).

[0057] (3)

[0058] in, The scale is Gaussian pyramid image, 、 Indicates that the image is at scale The second-order partial derivative under , Indicates that the image is at scale The mixed derivatives under , the actual calculation is accelerated by the box filter (BoxFilter) approximation.

[0059] Specifically, the seam calculation unit first uses the Fast Library for Approximate Nearest Neighbors (FLANN) algorithm to match feature points. FLANN is a matching operation method based on Kd-tree. Kd-tree is a balanced binary tree, which is mainly used to divide the K-dimensional data information distribution space and can quickly find high-dimensional data information. It is mainly aimed at the repeated use of super-distribution planes with mutually perpendicular distribution coordinate axes, so that the data information distribution space is divided into two major components. The steps to construct Kd-tree are:

[0060] 1. Determine the partition domain. Calculate the variance of each dimension in the data space, and the partition domain represents the dimension with the largest variance.

[0061] 2. Divide the space. Sort the calculated data from small to large according to the partition domain. After finding the middle node, divide the data space into two subspaces according to the partition domain.

[0062] 3. Continue recursively until the subspace contains no elements. If the subspace contains only a single element, partition it based on the first dimension of the element.

[0063] Specifically, the FLANN algorithm first analyzes the generated Kd-tree, then generates an index structure for the corresponding data information. Based on this data structure, feature point search is primarily based on comparing the Euclidean distance of the point, i.e., finding the closest Euclidean distance to the query point. Matching point pairs are determined based on the ratio of the nearest neighbor to the next nearest neighbor. After feature point matching, the overlapping area is obtained, and dynamic programming is used to find the optimal seam line in the overlapping area. The steps for constructing the seam line are as follows:

[0064] 1. Initialize the energy matrix. Each column of pixels has a corresponding seam line. Starting from the first row of pixels and continuing to the last row, we can get the standard value of the pixels corresponding to the seam line.

[0065] 2. Locate the direction of the stitching line. Starting from the second row, compare the stitching line energy values ​​of the current pixel with the three adjacent pixels in the next row, as shown in Equation (4).

[0066] (4)

[0067] in, is the lowest energy value of the current pixel passing through the seam, Indicates the energy value of the current pixel. At the same time, the corresponding upstream pixels are recorded, and the minimum criterion value point is regarded as the direction of the stitching line.

[0068] 3. Repeat the calculation of the second step until the last row of pixels is calculated, and take the one with the smallest criterion value as the best stitching line.

[0069] Specifically, in step S2, the video frame fusion module, after obtaining the optimal stitching line calculated by the frame stitching parameter calculation module, quickly stitches the subsequent video frames of the multiple video streams together into a subsequent wide-angle video stream based on the stitching line. Using the information obtained from the initialization template frame omits many steps in the subsequent video frame stitching, significantly reducing video stitching time.

[0070] Specifically, the embodiment of the present invention also provides an intelligent focusing module, including an attention recognition module, a visual blind spot calculation module, and an auxiliary acceleration sensor, such as Figure 3 As shown in the figure, the attention recognition module inputs the preprocessed image or video stream information, and outputs the key area recognition results in the image that simulates the visual characteristics of the human eye. By analyzing the image texture and visual blind spot characteristics, the areas that need priority attention are located.

[0071] Specifically, gaze point correction and attention recognition are first performed. The visual area is divided into 48 areas (6 rows x 8 columns), with a calibration point set at the center of each area. The test subject looks at these calibration points in turn, and the system records the gaze point error at each point. The error is decomposed into X-axis and Y-axis components, and the error value of each calibration point is calculated separately. A two-dimensional error surface is constructed using radial basis function (RBF) interpolation to dynamically correct the gaze point in any visual area.

[0072] Specifically, the reflection positions of four infrared LED spots on the cornea ( ) and its projection ( ), the pupil center is determined by the cross ratio invariance. The cross ratio is defined as shown in formula (5), which ensures the stability of the geometric relationship of the spot projection on the corneal surface.

[0073] (5)

[0074] Among them, | |Expressed as the absolute value of the distance between two points.

[0075] Specifically, the error caused by corneal non-planarity is corrected. The actual spot position ( ) is mapped to the virtual plane position ( ), as shown in formula (6).

[0076] (6)

[0077] in, represents the reference position on the retina (i.e. the intersection of the optical axis of the human eye and the retina), Through the pupil center The condition for coincidence with the virtual light spot is calculated as shown in formula (7).

[0078] (7)

[0079] Specifically, the error between the actual and predicted line-of-sight points is measured at each calibration point, decomposed into X / Y components, and vector correction is performed. A global error surface is constructed using Gaussian radial basis function interpolation, as shown in Equation (8). The interpolation function is a linear combination of the basis functions, as shown in Equation (9).

[0080] (8)

[0081] (9)

[0082] Determine weights by solving a system of linear equations Realize error compensation at any position.

[0083] Specifically, in step S3, the blind spot calculation input is the key area identification result, and the output is a dynamically adjusted detection weight allocation scheme. This automatically enhances attention to blind spot targets based on the real-time scene complexity, ensuring that the system can still accurately capture important targets in complex environments. A convolutional neural network is used to calculate key areas in the image and allocate attention to those areas that may contain important targets. The blind spot identification weights output by this module can be expressed as shown in Equation (10).

[0084] (10)

[0085] in, is the image point The visual blind spot identification weight, is the visual attention response computed by the network.

[0086] Specifically, the embodiment of the present invention also provides intelligent blind spot pedestrian detection. Through the intelligent blind spot matching strategy, the calculation path of the deep learning network model is optimized, and the detection weights of the left and right blind spots are adjusted according to the left and right acceleration information during the real-time detection process. Figure 4 As shown, the intelligent blind spot pedestrian detection module includes a blind spot weight evaluation module and a dynamic resource allocation module.

[0087] Specifically, in step 4, the blind spot weight assessment module takes as input the blind spot identifier and blind spot reinforcement identifier output by the intelligent focusing module. Combined with the driver's position information, it performs real-time analysis to output the weighted ratio for calculating the left and right blind spots. This module's inputs are two key factors: first, the blind spot identifier, which indicates the possible blind spots within the left and right blind spots and is typically related to the vehicle's dynamic state; and second, the blind spot reinforcement identifier, which indicates which areas should be strengthened, typically based on risk assessment and environmental dynamics. The general formula for weight calculation is shown in Equation (11).

[0088] (11)

[0089] in, and are the calculation weights for the left blind area and the right blind area, and They are the relevant area signs of the left and right blind spots, and The left and right acceleration information are analyzed in real time to assess the weighting of the left and right blind spots. This step helps determine the importance of each blind spot and prepares for subsequent resource allocation.

[0090] Specifically, in order to increase the complexity of the model, a nonlinear function containing more parameters can be used. The final weight function is expanded as shown in Equations (12) and (13).

[0091] (12)

[0092] (13)

[0093] in, , , , is the weight coefficient of the blind area identification, , It is a coefficient related to acceleration, which is used to reflect the impact of acceleration on weight. and The term in the form of indicates that the weight decreases exponentially as the acceleration increases, which is consistent with the smaller acceleration on the inner side of the turn in actual physical situations.

[0094] Specifically, assuming that the left blind spot area is marked , right blind spot area marking , left acceleration , right acceleration This indicates that there are more pedestrian targets in the left blind spot and the vehicle is turning left. Let’s assume that the driver is driving on the right side of the vehicle and the driver’s seat is on the left side. The driver pays more attention to the left side and thinks that there are more pedestrian targets. Assumptions , , , , , , then:

[0095] (14)

[0096] (15)

[0097] Therefore, it is believed that sudden traffic accidents are more likely to occur on the left side.

[0098] Specifically, in step 4, the dynamic resource allocation module outputs dynamically adjusted pedestrian detection results based on the computational weights of the left and right blind spots. This module intelligently allocates computing resources based on the weights of the left and right blind spots, concentrating computing resources on the blind spots most likely to cause accidents during the detection process. The resource allocation formula is shown in Equation (16).

[0099] (16)

[0100] in, is the final resource allocation, and are the computing resources required for the left and right blind areas respectively.

[0101] Specifically, by rationally allocating computing resources, the system ensures accurate detection of dangerous areas and avoids resource waste. Under the constraints of limited computing power, this module enables the system to balance accuracy and efficiency in real-time processing and quickly pass the detection results to the warning module in step S5. Figure 5 As shown, based on the results of intelligent blind spot pedestrian detection, the dashboard warning and audio warning are controlled to remind the driver of the "ghost peeking" situation, and the driver needs to brake immediately and pay attention to the surrounding situation.

[0102] Specifically, embodiments of the present invention provide an ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing, aiming to significantly improve pedestrian perception in complex dynamic scenes and effectively reduce traffic accidents caused by "ghost peeking." The detection method primarily consists of a preprocessing module, a blind spot perspective fusion module, an intelligent focusing module, an intelligent blind spot pedestrian detection module, and a driver warning module. The blind spot perspective fusion module, drawing inspiration from bionic structures in nature, employs an optimized strategy of multi-perspective fusion and distortion correction to overcome the perspective limitations of traditional visual detection systems, maintaining superior detection accuracy in an ultra-wide-angle field of view, exceeding conventional viewing angles. A distortion correction algorithm specifically designed for vehicle blind spots ensures stable detection under varying lighting conditions and complex scenarios. Embodiments of the present invention also incorporate an intelligent focusing module, based on optimization of human visual characteristics and an intelligent focusing mechanism using left and right accelerometers. This mechanism dynamically adjusts target detection weights to mimic the focusing characteristics of the human eye. Combined with data input from the left and right accelerometers, the system can perceive the vehicle's motion in real time and prioritize pedestrian features in potentially dangerous areas during steering movements. When the intelligent pedestrian detection module detects a pedestrian in the blind spot, the system immediately notifies the driver through the driver warning module, via an instrument panel warning and an audio alarm, prompting them to apply emergency braking. This embodiment of the present invention overcomes the limitations of traditional pedestrian detection systems in wide-angle fields and high-dynamic environments, significantly improving the robustness of target perception and detection. It has broad application prospects in areas such as autonomous driving and intelligent security.

[0103] The implementation of each embodiment of the present invention is based on programmed processing performed by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functionalities of each embodiment of the present invention are packaged into various modules. Based on this reality, and in addition to the aforementioned embodiments, an embodiment of the present invention provides an ultra-wide-angle pedestrian detection system based on perspective fusion and intelligent focusing. This system is used to implement the ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing described in the aforementioned method embodiments.

[0104] The system includes: a data acquisition module, which is used to obtain ultra-wide-angle pedestrian data to be tested, and perform preprocessing and blind spot perspective fusion; a pedestrian detection module, which is used to input the fused data into a trained pedestrian detection model to obtain pedestrian detection results; wherein, the training of the pedestrian detection model includes: constructing an ultra-wide-angle pedestrian data set; building a pedestrian detection model, including: an intelligent focusing module, which is used to add visual blind spot identification to the input data, and dynamically adjust the detection weights of the left and right blind spots with the assistance of an acceleration sensor; an intelligent blind spot pedestrian detection module, which is used to optimize the calculation path through an intelligent blind spot matching strategy, and adjust the detection weights of the left and right blind spots according to the left and right acceleration information during real-time detection; a driver warning module, which is used to remind the driver through instrument panel warnings and audio warnings based on the intelligent blind spot pedestrian detection results; and the constructed ultra-wide-angle pedestrian data set is used to train the constructed pedestrian detection model to obtain a trained pedestrian detection model.

[0105] The ultra-wide-angle pedestrian detection system based on perspective fusion and intelligent focusing provided by the embodiment of the present invention addresses the limitations of traditional pedestrian detection systems in wide-angle fields of view and high-dynamic environments. It adopts several modules, including a preprocessing module, a blind spot perspective fusion module, an intelligent focusing module, an intelligent blind spot pedestrian detection module, and a driver warning module. It breaks through the limitations of traditional pedestrian detection systems in wide-angle fields of view and high-dynamic environments, significantly improves the robustness of target perception and detection, and has broad application prospects in the fields of autonomous driving, intelligent security, etc.

[0106] Based on the same inventive concept as the aforementioned embodiment, an embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing proposed in the aforementioned embodiment.

[0107] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program. When executed by a processor, this program overcomes issues such as viewing angle limitations, insufficient distortion correction, and poor adaptability to dynamic scenes. It significantly improves the robustness of target perception and detection, transcending the viewing angle limitations of traditional visual detection systems. It enhances detection accuracy in ultra-wide-angle fields of view, enables real-time perception of vehicle motion, and prioritizes pedestrian features in potentially hazardous areas during steering movements.

[0108] The storage medium can be any non-volatile storage device such as a hard disk, solid-state drive, flash drive, optical disk, etc., which is used to store computer program code and necessary data files. The stored computer program includes: a data acquisition module and a pedestrian detection module.

[0109] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and is susceptible to numerous variations. Any simple modifications, equivalent variations, and modifications to the above specific embodiments based on the technical essence of the present invention shall be deemed to fall within the scope of protection of the present invention.

Claims

1. Ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing, characterized by: include: Obtain ultra-wide-angle pedestrian data to be tested, perform pre-processing and blind spot perspective fusion; The fused data is input into a trained pedestrian detection model to obtain a pedestrian detection result; wherein the training of the pedestrian detection model includes: Build an ultra-wide-angle pedestrian dataset; Build a pedestrian detection model, including: an intelligent focusing module, which adds visual blind spot markers to input data and dynamically adjusts the detection weights of the left and right blind spots using acceleration sensors; an intelligent blind spot pedestrian detection module, which optimizes the calculation path through an intelligent blind spot matching strategy and adjusts the detection weights of the left and right blind spots based on left and right acceleration information during real-time detection; and a driver warning module, which alerts the driver based on the results of the intelligent blind spot pedestrian detection. The constructed ultra-wide-angle pedestrian dataset is used to train the constructed pedestrian detection model to obtain a trained pedestrian detection model.

2. The ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing according to claim 1, characterized in that: Perform preprocessing and blind area perspective fusion, including: Preprocess the ultra-wide-angle pedestrian data, including denoising, normalization, and resizing; The pre-processed data is input into the blind spot view fusion module for splicing and fusion to obtain a large-view video stream.

3. The ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing according to claim 1, characterized in that: The intelligent focusing module further includes: Attention recognition module, used to analyze image texture and visual blind spot characteristics, and locate the visual blind spots that require priority attention; Visual blind spot calculation module, used to automatically enhance attention to targets in visual blind spots based on real-time scene complexity; The auxiliary acceleration sensor module is used to capture the acceleration on the left and right sides of the vehicle based on the vehicle's dynamic behavior and add enhanced identification to the left and right blind spot video streams.

4. The ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing according to claim 1, characterized in that: The intelligent blind spot pedestrian detection module further includes: The blind spot weight assessment module is used to combine the driver's position information to analyze and output the weight ratio of the left and right blind spots in real time; The dynamic resource allocation module is used to intelligently adjust internal computing resources based on the weight ratio of the left and right blind spots and reasonably allocate limited computing power.

5. The ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing according to claim 3 is characterized in that: The identification weight formula for calculating the visual blind spot is: , in, is the image point The visual blind spot identification weight, is the visual attention response computed by the network.

6. The ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing according to claim 2, characterized in that: The blind spot perspective fusion module further includes: Frame stitching parameter calculation module, used to extract feature points and calculate the optimal stitching line; The video frame fusion module is used to quickly stitch together a large-viewing angle video stream based on the optimal stitching line.

7. Ultra-wide-angle pedestrian detection system based on perspective fusion and intelligent focusing, characterized by: include: The data acquisition module is used to obtain the ultra-wide-angle pedestrian data to be tested, perform pre-processing and blind spot perspective fusion; The pedestrian detection module is used to input the fused data into the trained pedestrian detection model to obtain pedestrian detection results. The training of the pedestrian detection model includes: Build an ultra-wide-angle pedestrian dataset; Build a pedestrian detection model, including: an intelligent focusing module, which adds visual blind spot markers to input data and dynamically adjusts the detection weights of the left and right blind spots using acceleration sensors; an intelligent blind spot pedestrian detection module, which optimizes the calculation path through an intelligent blind spot matching strategy and adjusts the detection weights of the left and right blind spots based on left and right acceleration information during real-time detection; and a driver warning module, which alerts the driver based on the results of the intelligent blind spot pedestrian detection. The constructed ultra-wide-angle pedestrian dataset is used to train the constructed pedestrian detection model to obtain a trained pedestrian detection model.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing are implemented as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the ultra-wide-angle pedestrian detection method based on perspective fusion and intelligent focusing are implemented as described in any one of claims 1 to 6.

Images