A projection system for displaying a picture of a blind area of an A-pillar of a vehicle

Abstract

The application relates to the technical field of automobile A-pillar blind area display, and specifically provides a projection system for displaying an automobile A-pillar blind area picture, which comprises two cameras, is arranged on a left rearview mirror and a right rearview mirror outside a vehicle respectively, and is used for collecting A-pillar peripheral environment images; a three-color micro-vibration mirror projector arranged in the vehicle and used for projecting a picture obtained according to the environment images onto two A-pillars; a damping connection assembly used for connecting the three-color micro-vibration mirror projector on the top of the vehicle; an SMA collision buffering mechanism arranged on the damping connection assembly and used for pulling the three-color micro-vibration mirror projector away from the vehicle personnel when the vehicle collides; and a control unit used for carrying a joint perception prediction model, processing the environment images collected by the cameras through the joint perception prediction model, and projecting through the three-color micro-vibration mirror projector. The application does not change the A-pillar structure, is simple to deploy, and fully guarantees the safety of the driver by pulling the projector through the SMA spring.

Description

Technical Field

[0001] This invention relates to the field of automotive A-pillar blind spot display technology, specifically providing a projection system for displaying images of the automotive A-pillar blind spot. Background Technology

[0002] With the rapid iteration of the automotive industry, consumers are paying increasing attention to driving safety, and vehicle visibility safety has become a core aspect that cannot be ignored in the overall vehicle safety system. Among them, the blind spot caused by the A-pillar is one of the typical pain points affecting driving safety.

[0003] As a key component of the vehicle body structure, the A-pillar is a load-bearing structure connecting the front compartment and the roof on both sides of the driver's and passenger's seats. Its core function is to maintain the integrity of the vehicle's passenger compartment structure and provide sufficient survival space for occupants in the event of a collision, especially a rollover or frontal offset collision. It is a core structural component for ensuring the passive safety of the entire vehicle. However, due to limitations in vehicle body styling and structural strength design, the A-pillar inevitably obstructs the driver's view. When the vehicle is turning, the A-pillar creates a blind spot, where pedestrians, non-motorized vehicles, or obstacles can easily fall into the blind spot and become unseen by the driver, increasing the probability of a collision.

[0004] To address the A-pillar blind spot issue, various solutions have been developed. One approach involves using an external depth camera to capture real-time images of the blind spot and then displaying them to the driver via a device inside the A-pillar, effectively "seeing through" the obscured area. However, these existing technologies have several drawbacks in practical application. Firstly, to install a flexible LCD screen inside the A-pillar, automakers must either modify the original internal structure and materials of the A-pillar or add additional screen components. The former directly damages the A-pillar's load-bearing structure, reducing the vehicle's overall collision strength. The latter adds extra rigid protrusions, which can easily impact the driver's head during a frontal collision, posing a serious risk of secondary injury. Traditional rigid installations or simple, brittle plastic structures are completely inadequate for occupant protection in collision scenarios. Secondly, the material and structural modification costs of the entire flexible LCD screen solution remain high, making it difficult to widely adopt in low- to mid-range mass-produced vehicles and extend the A-pillar-seeing safety feature to a broader consumer base. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a projection system for displaying the blind spot image of a car's A-pillar. The system uses a camera in conjunction with a projector to project the blind spot image, and employs the inherent properties of an SMA spring to pull the projector, ensuring personnel safety. The system is simple to deploy, easy to implement, and highly safe.

[0006] The projection system for displaying the blind spot image of a car's A-pillar, provided by this invention, includes: Two cameras are mounted on the left and right rearview mirrors of the vehicle, respectively, to capture images of the environment surrounding the A-pillar; The three-color micro-mirror projector is installed inside the vehicle to project images obtained from environmental images onto the A-pillars on both sides. Vibration damping connection assembly for connecting the three-color micro-mirror projector to the roof of the vehicle; The SMA collision buffer mechanism, installed on the vibration damping connection assembly, is used to pull the three-color micro-mirror projector away from the occupants of the vehicle when a collision occurs. The control unit is used to carry a joint perception prediction model, which processes environmental images captured by the camera and projects them through a three-color micro-mirror projector.

[0007] Preferably, the housing of the three-color micro-mirror projector is provided with heat dissipation fins for heat dissipation of the three-color micro-mirror projector.

[0008] Preferably, the vibration damping connection assembly includes a vibration damping pad and a vibration damping bracket, wherein the vibration damping bracket includes an upper end face, a lower end face, and a flexible wire between the upper end face and the lower end face; The vibration damping pad is placed between the upper surface and the roof; the SMA collision buffer mechanism is connected to the lower surface.

[0009] Preferably, the SMA collision buffer mechanism includes a fixed base, a hinged bracket, and a shear pin; The fixed base is set on the lower end face, and the two ends of the hinge bracket are respectively hinged to the fixed base and the three-color micro-mirror projector. The two ends of the shear pin are respectively connected to the three-color micro-mirror projector and the fixed base; the three-color micro-mirror projector is supported by the hinge bracket and the shear pin.

[0010] Preferably, the SMA collision buffer mechanism also includes an SMA spring, with the two ends of the SMA spring connected to a fixed base and a three-color micro-mirror projector, respectively; under normal conditions, the SMA spring is in a stretched state.

[0011] Preferably, when a car collision occurs, the ambient temperature reaches the austenitic phase transformation end temperature of the SMA, the SMA spring actively contracts, breaks the shear pin, and pulls the three-color micro-mirror projector to rotate around the hinge end of the hinge bracket and move towards the roof.

[0012] Preferably, an energy absorption block is provided on the fixed base, which serves as a buffer for the three-color micro-mirror projector when it is moved.

[0013] Preferably, the improved CNN convolutional model is used to obtain the joint perception prediction model. The improvements include: replacing the traditional convolutional units in the encoder with a local-global context fusion module; and introducing a feature alignment correction mechanism in the decoupled perception head.

[0014] Preferably, the process of projection through the joint perception prediction model is as follows: S1: Input the normalized environment image, aggregate static spatial context information using large kernel convolution operation to obtain static guiding tensor; use the original features as query vector, concatenate with the static guiding tensor in channel dimension, and generate dynamic attention weight matrix through continuous nonlinear transformation; weight and aggregate the dynamic attention weight matrix with the original value vector to generate enhanced feature map; S2: Input the enhanced feature map into the parallel multi-task decoupled perception head for split processing: the detection branch directly outputs the target's center point heatmap, bounding box size, and sub-pixel-level center offset to obtain the target's preliminary localization bounding box; the feature alignment correction mechanism is used to migrate the sampling position of the re-identification branch to the core region that distinguishes the target; the parallel re-identification branch extracts the identity embedding vectors used to distinguish different targets from the enhanced feature map; S3: Based on the target's initial localization bounding box and identity embedding vector, construct a geometric adaptive state estimator based on Kalman filtering. Use a uniform motion model as the state transition equation to predict the target's prior position and geometric shape in the next frame, thus forming the target trajectory. S4: In the data association stage of the joint perception prediction model, a hierarchical confidence matching strategy is adopted to match the newly identified target to the corresponding target trajectory, thereby achieving target spatial location matching and obtaining target tracking results. S5: Call the intrinsic parameter matrix of the three-color micro-mirror projector and the geometric correction parameters of the inner curved surface of the A-pillar. Based on the target tracking results, map the two-dimensional bounding box coordinates of the target to the three-dimensional virtual projection plane. Generate an augmented reality cursor with directional guidance and a highlighted outline through the rendering engine to obtain the image stream after surface adaptation and correction. The image stream drives the micro-mirror projection module of the three-color micro-mirror projector to project the synthesized image onto the inner side of the A-pillar in real time.

[0015] Preferably, the hierarchical cascaded confidence matching strategy includes two levels of association, specifically: Based on the confidence threshold, targets exceeding the confidence threshold are classified as high-confidence targets, and the rest are low-confidence targets. In the first-level association, the fusion distance matrix between high-confidence targets and matched historical target trajectories is calculated, and the Hungarian algorithm is used to complete the identity matching of high-confidence targets. In the second-level association, the unmatched historical target trajectories are matched with the low-confidence targets in terms of spatial location.

[0016] Compared with the prior art, the present invention can achieve the following beneficial effects: This invention requires no structural modifications to the A-pillar, fundamentally ensuring the overall structural strength and passive safety of the vehicle. Furthermore, by projecting real-time images of the external environment, it completely eliminates the blind spot created by the A-pillar, significantly improving active safety during daily driving. Compared to existing technologies, this solution avoids the high cost of traditional display screens, offering lower implementation costs and easier maintenance. Simultaneously, its overall structure occupies less space, minimizing encroachment on valuable interior space and eliminating the issue of additional components obstructing the driver's view.

[0017] The SMA anti-collision buffer mechanism of this invention is entirely controlled by a purely mechanical structure, completely avoiding the response delay problem caused by the stacking of multiple algorithm programs. Relying on the inherent properties of SMA material, it can directly trigger action when the ambient temperature or collision inertial force reaches a preset threshold, and the operation process is stable and reliable. At the same time, this invention introduces a joint learning framework integrating perception, tracking, and prediction. Through feature sharing and collaborative optimization mechanisms, the perception features provide richer scene context information for tracking and prediction tasks. The feedback from the tracking and prediction modules can also improve the recognition accuracy of perception in difficult scenarios such as rain, fog, and occlusion. This fundamentally avoids the common problem in traditional independent module pipelines where "upstream errors directly lead to downstream system crashes". The final output prediction result no longer relies solely on a simple kinematic model, but integrates deep scene semantic information, which can accurately predict complex interaction behaviors between vehicles and pedestrians. The generated driving trajectory prediction results are more reasonable and closer to real road conditions.

[0018] By relying on a unified high-dimensional feature representation, the system of this invention achieves a high degree of information consistency in the time dimension, significantly reducing the frequent ID switching problems in traditional tracking schemes and significantly improving the overall robustness of the system. Compared with three independently deployed large models, this invention significantly reduces the overall computational load and memory usage. At the same time, it eliminates the redundant steps of data serialization, transmission, and deserialization between different modules in the traditional pipeline, directly passing high-dimensional feature tensors within the model. This results in a qualitative leap in system operating efficiency and extremely low latency, making it more suitable for driving warning systems that require real-time projection of the external environment. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a camera view provided according to an embodiment of the present invention; Figure 2 This is a partial flowchart of the joint sensing prediction model provided according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and do not constitute a limitation thereof. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined to form various implementations. Furthermore, the order of the steps or actions in the method description can be changed or adjusted in a manner readily apparent to those skilled in the art. Therefore, the various orders in the specification and drawings are merely for the clear description of a particular embodiment and do not imply a mandatory order, unless otherwise stated that a particular order must be followed.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention 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 on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0023] In the description of this invention, it should be noted that, unless otherwise explicitly 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 will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0024] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] This invention provides a projection system for displaying the blind spot image of a car's A-pillar. The system includes two cameras, a three-color micro-mirror projector, a vibration-damping connection assembly, an SMA (Surface Mount Massive Amplifier) ​​collision buffer mechanism, and a control unit. The two cameras are respectively mounted above the two rearview mirrors on the left and right sides of the vehicle, used to acquire environmental images of the area surrounding the A-pillars in real time, frame by frame. The three-color micro-mirror projector is mounted on the roof inside the vehicle, used to project the processed image onto the A-pillars. The vibration-damping connection assembly is located between the three-color micro-mirror projector and the roof, used to connect the projector and simultaneously dampen its vibration. The SMA collision buffer mechanism, mounted on the vibration-damping connection assembly, pulls the projector in the event of an accident to protect the safety of the occupants. The control unit is electrically connected to the other structures, used to control them to perform corresponding operations, process the acquired environmental images, and project the processed image onto the A-pillars via the three-color micro-mirror projector, thus achieving the projection of the blind spot image.

[0026] The three-color micro-mirror projector's casing is equipped with heat dissipation fins for cooling. The vibration damping connection assembly includes damping pads and a damping bracket. The damping pads are positioned between the damping bracket and the vehicle roof to absorb high-frequency vibrations, while the damping bracket absorbs low-frequency, large-amplitude shaking. The damping bracket includes an upper end face, a lower end face, and multiple sets of flexible wires positioned between the upper and lower end faces. The upper end face is fixedly connected to the vehicle's interior roof. The flexible wires allow the lower end face to move slightly relative to the upper end face, thereby absorbing low-frequency vibrations.

[0027] The SMA collision buffer mechanism is located on the upper end face and its structure includes a fixed base, hinged brackets, an SMA spring, a shear pin, and an energy absorption block. The fixed base is located on the lower end face. The two ends of the hinged brackets, SMA spring, and shear pin are respectively connected to the fixed base and the three-color micromirror projector. Specifically, there are multiple hinged brackets; this embodiment uses two. The two ends of each hinged bracket are hinged to the fixed base and the three-color micromirror projector, respectively. When only the hinged brackets are connected, the three-color micromirror projector can be rotatably connected to the fixed base via the hinged brackets. The two ends of the shear pin are respectively connected to the fixed base and the three-color micromirror projector. Under normal conditions, the two hinged brackets and one shear pin form an approximate tripod support structure, supporting the three-color micromirror projector and keeping it stationary. The two ends of the SMA spring are respectively connected to the fixed base and the three-color micromirror projector, and the SMA spring is in a stretched state. The contraction force of the SMA spring is much greater than the design shear force of the shear pin.

[0028] When a vehicle experiences a severe collision, the interior temperature rises to the austenitic transformation end temperature of the SMA (this temperature is below the ignition point of interior components but above the temperature of normal sun exposure). The SMA spring then actively contracts, breaking the shear pin, which in turn pulls the tri-color micro-mirror projector to rotate around the hinge end of the hinged bracket and move towards the roof to move it away from the occupants. If, during a severe collision, the temperature rise does not reach the austenitic transformation end temperature of the SMA, the collision will still generate a huge inertial force. This huge inertial force will directly shear the shear pin, which acts as a mechanical safety device. In this case, although the SMA spring does not actively contract, it no longer forms a rigid connection, and the tri-color micro-mirror projector can swing under inertia, similarly reducing intrusion.

[0029] An energy absorption block is installed on the fixed base. The energy absorption block is located at the end of the moving trajectory of the three-color micro-mirror projector. When the three-color micro-mirror projector is pulled to move, the energy absorption block is used to buffer the three-color micro-mirror projector.

[0030] The control unit is equipped with a joint perception prediction model, which processes multi-frame environmental images captured in real time by the camera. The joint perception prediction model is an improvement on the CNN convolutional model, with improvements including: replacing traditional convolutional units in the CNN convolutional model encoder with a local-global context fusion module; and introducing a feature alignment correction mechanism in the decoupled perception head of the CNN convolutional model. The process of the joint perception prediction model processing environmental images is as follows: S1: Environmental images contain complex backgrounds, such as roadside trees and building textures. These backgrounds are easily confused with targets such as pedestrians, therefore, target extraction and feature enhancement are necessary. This invention introduces a local-global context fusion module in the encoder of the joint perception prediction model to replace the traditional convolutional units in the CNN convolutional model encoder. The local-global context fusion module abandons the isolated pixel feature extraction method and instead constructs local neighborhood windows on the feature map. Specifically: The environmental image is normalized and then input into a joint perception prediction model. Static spatial context information is aggregated using large-kernel convolution operations to obtain a static guiding tensor. The original features are used as query vectors and concatenated with the static guiding tensor along the channel dimension. A dynamic attention weight matrix is ​​generated through continuous nonlinear transformations, explicitly encoding the semantic association strength between adjacent pixels. The dynamic attention weight matrix is ​​then weighted and aggregated with the original value vectors to generate an enhanced feature map, effectively suppressing background noise.

[0031] S2: After obtaining the enhanced feature map, it is input into a parallel multi-task decoupled perception head for split processing: the detection branch directly outputs the target's center point heatmap, bounding box size, and sub-pixel-level center offset to obtain the target's preliminary localization bounding box. To address potential pose changes of the target from the A-pillar camera's perspective, the re-identification branch is designed as a parallel structure. Since downsampling operations in convolutional neural networks may cause misalignment between the feature center and the geometric center, this embodiment introduces a feature alignment correction mechanism. Using this mechanism, an adaptive sampling offset field is learned, shifting the sampling position of the re-identification branch to the core region that distinguishes the target. Through the parallel-running re-identification branch, identity embedding vectors used to distinguish different targets are extracted from the enhanced feature map. The feature alignment correction mechanism ensures that even under conditions of vehicle vibration or non-rigid deformation of the target, the extracted identity embedding vectors maintain high consistency and separability.

[0032] S3: To achieve smooth and accurate target tracking in the projected image, this embodiment of the invention constructs a geometrically adaptive state estimator based on Kalman filtering, using the initial target localization bounding box and identity embedding vector. Traditional aspect ratio-based state descriptions often suffer from bounding box jitter when faced with rapid shape changes caused by pedestrians stepping or vehicles turning. This embodiment of the invention defines the target's state vector as... The width of the target 2D bounding box and height Independent regressions are performed on the target and its rate of change. Using a uniform motion model as the state transition equation, the prior position and geometry of the target in the next frame are predicted, forming the target trajectory.

[0033] The direct dimension regression strategy of this invention avoids the accumulation of errors caused by aspect ratio estimation, enabling the predicted box to fit more closely to the true edge of the target, and providing a reliable geometric benchmark for subsequent accurate projection.

[0034] S4: In this embodiment of the invention, a hierarchical confidence matching strategy is adopted in the data association stage of the joint perception prediction model to minimize the false negative rate of the A-pillar blind zone. Newly identified targets are matched to their corresponding target trajectories to achieve target spatial location matching and obtain target tracking results. The hierarchical confidence matching strategy includes two levels of association: according to a confidence threshold, targets exceeding the threshold are classified as high-confidence targets, and the rest are low-confidence targets; in the first level of association, the fusion distance matrix between high-confidence targets and matched historical target trajectories is calculated, and the Hungarian algorithm is used to complete the identity matching of high-confidence targets; in the second level of association, unmatched historical target trajectories are matched with low-confidence targets based on intersection-union ratio (IU).

[0035] The hierarchical confidence matching strategy can effectively retrieve the true target whose confidence has decreased due to environmental factors, maintain the continuity of the target in the A-pillar projection image, and avoid visual flickering interference when the driver observes.

[0036] S5: The final output stage of the joint perception prediction model performs the mapping from the perception space to the projection space. It calls the intrinsic parameter matrix of the three-color micro-mirror projector and the geometric correction parameters of the inner surface of the A-pillar, mapping the target's 2D bounding box coordinates to a 3D virtual projection plane based on the target tracking results. The rendering engine generates an augmented reality cursor with directional guidance and a highlighted outline, obtaining an image stream after surface adaptation and correction. This image stream drives the micro-mirror projection module of the three-color micro-mirror projector, projecting the synthesized image onto the inner side of the A-pillar in real time. This creates a "transparent" field of view on the inner side of the A-pillar that matches the real-world perspective outside the vehicle, thus physically eliminating the blind spot of the A-pillar.

[0037] In addition, to complement the system described in this application, the microstructure of the A-pillar interior panel is a diffuse reflection layer. Without altering the original material and color of the A-pillar, specific microprisms or diffuse reflection textures are formed on its surface through micro-embossing or laser engraving processes. This microstructure significantly enhances the visibility and contrast of the projected image under ambient light interference, while maintaining the original appearance of the A-pillar from the driver's perspective.

[0038] Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0039] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A projection system for displaying the blind spot image of a car's A-pillar, characterized in that, include: Two cameras are mounted on the left and right rearview mirrors of the vehicle, respectively, to capture images of the environment surrounding the A-pillar; A three-color micro-mirror projector is installed inside the vehicle to project images obtained from the environmental images onto the A-pillars on both sides. A vibration damping connection assembly is used to connect the three-color micro-mirror projector to the roof of the vehicle. The SMA collision buffer mechanism, installed on the vibration damping connection assembly, is used to pull the three-color micro-mirror projector away from the occupants of the vehicle when a collision occurs. The control unit is equipped with a joint perception prediction model, which processes the environmental images captured by the camera and projects them through the three-color micro-mirror projector.

2. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 1, characterized in that, The housing of the three-color micro-mirror projector is equipped with heat dissipation fins for heat dissipation.

3. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 1, characterized in that, The vibration damping connection assembly includes a vibration damping pad and a vibration damping bracket. The vibration damping bracket includes an upper end face, a lower end face, and a flexible wire between the upper end face and the lower end face. The vibration damping pad is disposed between the upper end surface and the roof; the SMA collision buffer mechanism is connected to the lower end surface.

4. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 3, characterized in that, The SMA collision buffer mechanism includes a fixed base, a hinged bracket, and a shear pin; The fixed base is disposed on the lower end surface, and the two ends of the hinge bracket are respectively hinged to the fixed base and the three-color micro-mirror projector. The two ends of the shear pin are respectively connected to the three-color micro-mirror projector and the fixed base; the three-color micro-mirror projector is supported by the hinge bracket and the shear pin.

5. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 4, characterized in that, The SMA collision buffer mechanism also includes an SMA spring, with its two ends connected to the fixed base and the three-color micro-mirror projector, respectively; under normal conditions, the SMA spring is in a stretched state.

6. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 5, characterized in that, When a car collision occurs, the ambient temperature reaches the austenitic phase transformation end temperature of the SMA. The SMA spring actively contracts, breaks the shear pin, and pulls the three-color micro-mirror projector to rotate around the hinge end of the hinge bracket and move towards the roof of the car.

7. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 6, characterized in that, An energy absorption block is provided on the fixed base. When the three-color micro-mirror projector is moved, the energy absorption block is used to buffer the three-color micro-mirror projector.

8. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 1, characterized in that, The joint perception prediction model is obtained by improving the CNN convolutional model. The improvements include: replacing the traditional convolutional units in the encoder with a local-global context fusion module; and introducing a feature alignment correction mechanism in the decoupled perception head.

9. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 8, characterized in that, The process of projection through the joint sensing prediction model is as follows: S1: Input the normalized environment image, aggregate static spatial context information using large kernel convolution operation to obtain a static guiding tensor; use the original features as query vectors, concatenate them with the static guiding tensor in the channel dimension, and generate a dynamic attention weight matrix through continuous nonlinear transformation; weight and aggregate the dynamic attention weight matrix with the original value vector to generate an enhanced feature map; S2: Input the enhanced feature map into the parallel multi-task decoupled perception head for split processing: the detection branch directly outputs the target's center point heatmap, bounding box size, and sub-pixel-level center offset to obtain the target's preliminary localization bounding box; The sampling position of the re-identification branch is moved to the core region that distinguishes the target by using a feature alignment correction mechanism; The parallel re-identification branch extracts identity embedding vectors from the enhanced feature map to distinguish different targets; S3: Based on the target's initial localization bounding box and identity embedding vector, construct a geometric adaptive state estimator based on Kalman filtering. Use a uniform motion model as the state transition equation to predict the target's prior position and geometric shape in the next frame, thus forming the target trajectory. S4: In the data association stage of the joint perception prediction model, a hierarchical confidence matching strategy is adopted to match the newly identified target to the corresponding target trajectory, thereby achieving target spatial location matching and obtaining target tracking results. S5: Call the intrinsic parameter matrix of the three-color micro-mirror projector and the geometric correction parameters of the inner curved surface of the A-pillar. Based on the target tracking results, map the two-dimensional bounding box coordinates of the target to the three-dimensional virtual projection plane. Generate an augmented reality cursor with directional guidance and a highlighted outline through the rendering engine to obtain an image stream after surface adaptation and correction. The image stream drives the micro-mirror projection module of the three-color micro-mirror projector to project the synthesized image onto the inner side of the A-pillar in real time.

10. The projection system for displaying the blind spot image of a car's A-pillar as described in claim 9, characterized in that, The hierarchical cascaded confidence matching strategy includes two levels of association, specifically: Based on the confidence threshold, targets exceeding the confidence threshold are classified as high-confidence targets, and the rest are low-confidence targets. In the first-level association, the fusion distance matrix between high-confidence targets and matched historical target trajectories is calculated, and the identity matching of high-confidence targets is completed using the Hungarian algorithm. In the second-level association, unmatched historical target trajectories are matched with low-confidence targets in terms of spatial location.

Images