Autonomous vehicle evidence investigation system based on panoramic image reconstruction and trace collection

Abstract

This invention relates to an autonomous vehicle-based evidence investigation system based on panoramic image reconstruction and trace acquisition, belonging to the field of forensic science technology. The system includes an intelligent mobile platform and integrated planar information acquisition modules, elevation information acquisition modules, spatial information acquisition modules, and darkroom information acquisition modules. The intelligent mobile platform serves as the supporting foundation, adaptable to complex terrain operations. The planar information acquisition module clearly captures two-dimensional traces on the ground and preserves their outlines. The elevation information acquisition module searches for and extracts traces on vertical and inclined surfaces. The spatial information acquisition module acquires panoramic images, which are then stitched and fused to generate a distortion-free panoramic image. Combined with positioning information and laser point cloud data, a three-dimensional model of the scene is constructed. The darkroom information acquisition module records evidence imaging in a darkroom setting. The system supports inter-module collaborative acquisition and data preprocessing, verification, and associated storage, improving the evidence detection rate and data reliability in complex environments.

Description

Technical Field

[0001] This invention belongs to the field of forensic science technology, specifically relating to an autonomous vehicle-based evidence investigation system based on panoramic image reconstruction and trace acquisition. Background Technology

[0002] In the field of forensic science, the examination of physical evidence in criminal cases is a core aspect of case investigation and courtroom examination. Its core requirements lie in the accurate extraction of traces at the scene (such as footprints, tool marks, and trace evidence), the complete recording of the spatial relationships of physical evidence, and ensuring the authenticity and reliability of the collected data. With the increasing complexity of crime scene environments, traditional methods of physical evidence examination have gradually revealed significant technical limitations, specifically in the following aspects:

[0003] Criminal crime scenes often involve dangerous restricted areas, high-risk areas, or enclosed dark rooms. During manual investigation, investigators must directly enter dangerous environments, facing safety threats such as poisoning, falls, and structural collapses. Traditional wheeled mobile equipment is limited by terrain and cannot flexibly reach investigation points. In dark room environments, insufficient lighting and high imaging noise result in poor trace collection quality.

[0004] Physical evidence at the scene is distributed across multiple dimensions, including "plane (ground), vertical (walls, cabinet sides), and space (overall environment)," and some traces (such as dust scratches and traces of grease) require specific light sources to be revealed. During manual investigation, it is difficult to collect evidence from all areas without blind spots due to limitations in perspective and physical strength. Traditional equipment often focuses on a single dimension and has fixed light source functions, resulting in the omission of a large number of potential traces and affecting the efficiency of case investigation.

[0005] The spatial relationship of physical evidence is crucial for reconstructing a case. The location of physical evidence recorded manually relies on tape measure measurements and hand-drawn sketches, which are prone to positional deviations due to human error. The image data collected by traditional equipment (such as planar trace photographs and partial panoramic images) are independent of each other and lack a binding mechanism with the three-dimensional space of the scene. It is impossible to construct a "trace-spatial environment" relationship, making it difficult to reconstruct the scene logic of the formation of physical evidence during subsequent analysis, and also failing to provide an intuitive spatial evidence chain for court examination.

[0006] Manual surveying requires manually organizing photos and measuring data, which is time-consuming and labor-intensive. Data collected by traditional equipment often lacks preprocessing mechanisms (such as not removing image noise or calibrating color deviations) and lacks data verification functions. If data is lost or images are blurred during the collection process, it is necessary to re-enter the site to collect data, which further increases the survey cost and time cost. Especially in dangerous sites, secondary data collection will face safety risks again.

[0007] Currently, there is a need for an investigation system that can adapt to special scenarios such as danger, darkrooms, and complex terrain, achieve multi-dimensional collaborative collection of traces from "planar-elevation-space-darkroom" perspectives, and can correlate the spatial location of physical evidence and ensure data reliability. This system would address core issues such as high safety risks of manual investigation, limited functionality of traditional equipment, and low data correlation, thereby improving the efficiency, accuracy, and security of physical evidence investigation in criminal cases. Summary of the Invention

[0008] To address the aforementioned problems in the existing technology, this invention provides an autonomous vehicle evidence investigation system based on panoramic image reconstruction and trace acquisition.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] An autonomous vehicle-based evidence investigation system based on panoramic image reconstruction and trace acquisition includes an intelligent mobile platform, a planar information acquisition module, an elevation information acquisition module, a spatial information acquisition module, and a darkroom information acquisition module.

[0011] The intelligent mobile platform is a mobile platform based on complex environments, carrying an information collection module;

[0012] The planar information acquisition module captures two-dimensional planar traces on the ground through the area array image acquisition component, eliminates ground shadow interference through diffused light source output, and the automatic focus adjustment component adjusts the acquisition focal length according to the ground distance to generate acquisition and outline preservation images of planar physical evidence.

[0013] The facade information acquisition module is based on a linear array scanning acquisition component. It scans the vertical or inclined surface of the facade line by line. A multi-band special light source component excites the fluorescent reaction or contrast imaging of the traces. An angle adjustment mechanism drives the linear array scanning acquisition component to rotate within a preset range to perform trace search and extraction.

[0014] The spatial information acquisition module acquires airborne harmful gas data and generates a gas concentration heat map. The circumferentially distributed lenses simultaneously acquire panoramic images of the scene. The images are seamlessly fused through an image stitching processing mechanism to generate a distortion-free panoramic image. The spatial coordinate association component combines the positioning information of the intelligent mobile platform to bind the panoramic image with the three-dimensional spatial location of the scene. Based on laser point cloud data, a texture mapping between the panoramic image and the point cloud model is generated to construct a three-dimensional model of the scene that includes the spatial location information of the physical evidence.

[0015] In a darkroom environment, the darkroom information acquisition module converts weak light signals into night vision images, supplements the light with non-visible light bands, and reduces the grain noise of the night vision images by the noise suppression processing component, thereby generating images and records of footprints, tool marks, and trace evidence in the darkroom scene.

[0016] Specifically, the area array image acquisition component divides the acquisition area into equal detection sub-regions and samples the ground reflected light intensity of the detection sub-regions; if the reflected light intensity of the detection sub-region exceeds a preset reflection threshold, a reduction command is sent to the diffuse light source; the area array image acquisition component acquires local test images in real time and generates the grayscale distribution uniformity of each sub-region in the image.

[0017] Specifically, the planar information acquisition module performs edge enhancement processing on the acquired planar trace image based on an image preprocessing mechanism. It adopts a direction enhancement algorithm based on gray-level gradient, which strengthens the gray-level difference of edge pixels by identifying the gray-level change trend of the trace edge. By extracting the main color information of the ground background, a background color template is constructed. Noise pixels in the planar trace image that deviate from the background color template but do not belong to the trace features are replaced with the gray-level values ​​of adjacent background pixels, thus preserving the color features of the trace itself. The preprocessed image is quality evaluated by calculating the contrast value between the trace area and the background area. If the contrast value reaches a preset threshold, the preprocessed image is marked and stored.

[0018] Specifically, during the process of the multi-band special light source component stimulating the trace reaction, a first-band light source is preset according to the matching relationship between the material of the facade and the type of trace; after the first-band light source is activated, the linear array scanning acquisition component scans the target facade area line by line and transmits the scanned image in real time; by calculating the grayscale variance of the image, it is determined that no potential trace is detected, and the system switches to the second-band light source and adjusts the scanning parameters of the linear array scanning acquisition component.

[0019] Specifically, after capturing physical evidence, the planar information acquisition module or the facade information acquisition module sends a trigger signal including the three-dimensional coordinates of the physical evidence and the type of physical evidence; after receiving the signal, it analyzes the coordinates of the physical evidence, sends a command to the multi-lens circular array panoramic camera, and adjusts the lens in the corresponding direction to align with the local area where the physical evidence is located; it adjusts the acquisition parameters of the panoramic camera, and after the acquisition is completed, it calculates the sharpness value of the local image.

[0020] Specifically, the image stitching processing mechanism extracts the overlapping area of ​​the lens images, selects color feature points within the overlapping area, calculates the color parameters of the feature points in the lens images using a color analysis algorithm, generates a color adjustment mapping table based on the color parameters of one lens, performs color correction on other lens images according to the color adjustment mapping table, and calculates the color difference between pixels on both sides of the stitching line.

[0021] Specifically, the spatial coordinate association component binds the panoramic image with the on-site three-dimensional spatial location, obtains the coordinate data of the current acquisition location, extracts the three-dimensional coordinates of fixed feature points on-site from the laser point cloud data, calculates the difference between the coordinate data and the three-dimensional coordinates, and calculates the deviation compensation value using a weighted average method; adjusts the coordinate binding parameters of the panoramic image according to the deviation compensation value, and corrects the coordinates of each pixel in the panoramic image to the original bound coordinates and the deviation compensation value.

[0022] Specifically, the process of supplementing light in the non-visible light band by the darkroom information acquisition module includes: scanning the darkroom environment to generate a global grayscale image, generating grayscale abnormal areas, and marking the boundary coordinates of suspected traces; sending a supplementing light adjustment command to the darkroom information acquisition module, wherein the supplementing light range in the non-visible light band covers the area enclosed by the boundary coordinates of the suspected traces.

[0023] Specifically, the noise suppression processing component reduces grain noise in night view images by comparing the night view images before and after noise suppression to locate the trace detail areas lost due to noise suppression; the detail restoration algorithm constructs a grayscale distribution model of the trace details based on the grayscale change trend of adjacent pixels around the lost area, and calculates the pixel grayscale values ​​of the lost area through interpolation; the restored image is compared with the original image before noise suppression, and if the deviation is less than a preset value, the restored image is stored.

[0024] Specifically, after the planar information acquisition module captures two-dimensional planar traces on the ground, it generates a planar trace image and sends a linkage trigger signal, including the trace acquisition timestamp, ground two-dimensional coordinates, and preliminary trace classification results; the two-dimensional coordinates of the trace are converted into spatial three-dimensional coordinates and sent to the spatial information acquisition module; after acquiring a panoramic image of the trace area, the planar trace image is associated and stored with the corresponding spatial panoramic image: the same association identifier code is assigned to both, the acquisition timestamp, acquisition location coordinates, and acquisition parameters are recorded, and an association index table is established.

[0025] Specifically, when the facade information acquisition module scans the facade in a low-light environment where the ambient light intensity is lower than a preset low-light threshold, it uses an image edge detection algorithm to identify suspected trace areas with abrupt changes in grayscale values ​​and discontinuous contours. When the image sharpness value is lower than a preset standard, it sends a linkage request signal, which includes the facade location coordinates and imaging parameters of the suspected trace area. The darkroom information acquisition module is then activated to perform secondary acquisition on the suspected trace area. After generating a night view image, the suspected trace image undergoes feature comparison.

[0026] Specifically, after the spatial information acquisition module constructs a three-dimensional model of the site, it annotates the airborne harmful gas data as semantic information to generate a gas concentration heat map; it marks the physical evidence locations in the three-dimensional model of the site, associates the corresponding planar acquisition images, elevation acquisition images, and darkroom acquisition images with the physical evidence, and generates a physical evidence association archive that includes multi-dimensional acquisition data.

[0027] The present invention, through the design of "intelligent mobile platform + multi-dimensional acquisition module + collaborative control mechanism", has the following specific beneficial effects:

[0028] Intelligent mobile platforms are adaptable to hazardous restricted areas, high-risk zones, and enclosed dark rooms, eliminating the need for survey personnel to enter directly. This fundamentally avoids safety threats such as poisoning, falls, and structural collapse, ensuring the safety of survey personnel. Compared to traditional wheeled equipment, intelligent mobile platforms have stronger terrain adaptability, enabling stable movement in complex terrains such as rugged ground and narrow passages. They can accurately reach survey points that are difficult for humans to access (such as crevices in ruins and corners of dark rooms), ensuring uninterrupted survey work in special scenarios.

[0029] The system integrates four major acquisition modules: planar, elevation, spatial, and darkroom, forming a comprehensive acquisition system covering "ground-elevation-overall space-low-light environment." The planar module captures footprints and dust scratches on the ground; the elevation module extracts bloodstains and grease marks on walls; the spatial module records the overall environmental correlation; and the darkroom module uncovers potential traces under low light conditions, ensuring no trace is missed. The spatial information acquisition module, through a spatial coordinate association component, binds panoramic images with the positioning information and laser point cloud data of the intelligent mobile platform, achieving a precise correspondence between "panoramic image - 3D spatial location - evidence coordinates," ensuring the accuracy of the recorded spatial location of evidence.

[0030] The system supports the coordinated acquisition of data from modules such as planar and spatial dimensions, and facade and darkroom. The acquired data is bound through "association identification code + index table" (e.g., associating planar trace images with corresponding spatial panoramic images and secondary acquisition images in the darkroom), constructing a complete data chain of "trace morphology - spatial environment - acquisition parameters". During subsequent analysis, the formation scene of physical evidence can be quickly traced, providing intuitive spatial logic support for case reconstruction. Attached Figure Description

[0031] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0032] Figure 1 This is an overall architecture diagram of an autonomous vehicle evidence investigation system based on panoramic image reconstruction and trace acquisition according to the present invention.

[0033] Figure 2 This is a diagram illustrating the module linkage mechanism in this invention. Detailed Implementation

[0034] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0035] Please see Figures 1-2 An autonomous vehicle-based evidence investigation system based on panoramic image reconstruction and trace acquisition includes an intelligent mobile platform, a planar information acquisition module, a facade information acquisition module, a spatial information acquisition module, and a darkroom information acquisition module.

[0036] The intelligent mobile platform is a mobile platform based on complex environments, carrying an information collection module;

[0037] The planar information acquisition module captures two-dimensional planar traces on the ground through the area array image acquisition component, eliminates ground shadow interference through diffused light source output, and the automatic focus adjustment component adjusts the acquisition focal length according to the ground distance to generate acquisition and outline preservation images of planar physical evidence.

[0038] The facade information acquisition module is based on a linear array scanning acquisition component. It scans the vertical or inclined surface of the facade line by line. A multi-band special light source component excites the fluorescent reaction or contrast imaging of the traces. An angle adjustment mechanism drives the linear array scanning acquisition component to rotate within a preset range to perform trace search and extraction.

[0039] The spatial information acquisition module acquires airborne harmful gas data and generates a gas concentration heat map. The circumferentially distributed lenses simultaneously acquire panoramic images of the scene. The images are seamlessly fused through an image stitching processing mechanism to generate a distortion-free panoramic image. The spatial coordinate association component combines the positioning information of the intelligent mobile platform to bind the panoramic image with the three-dimensional spatial location of the scene. Based on laser point cloud data, a texture mapping between the panoramic image and the point cloud model is generated to construct a three-dimensional model of the scene that includes the spatial location information of the physical evidence.

[0040] In a darkroom environment, the darkroom information acquisition module converts weak light signals into night vision images, supplements the light with non-visible light bands, and reduces the grain noise of the night vision images by the noise suppression processing component, thereby generating images and records of footprints, tool marks, and trace evidence in the darkroom scene.

[0041] Specifically, the area array image acquisition component divides the acquisition area into equal detection sub-regions and samples the ground reflected light intensity of the detection sub-regions; if the reflected light intensity of the detection sub-region exceeds a preset reflection threshold, a reduction command is sent to the diffuse light source; the area array image acquisition component acquires local test images in real time and generates the grayscale distribution uniformity of each sub-region in the image.

[0042] The area array image acquisition component divides the acquisition area into multiple equally sized detection sub-regions and samples the ground reflected light intensity of each sub-region in real time. If the reflected light intensity of a certain sub-region exceeds a preset reflection threshold (determined by the preset reflection characteristics of the ground material), the control module sends a reduction command to the diffuse light source. The light source gradually reduces the driving current of the light-emitting element to achieve a brightness gradient reduction, avoiding overexposure of trace details caused by strong light reflection. If the reflected light intensity of a certain sub-region is lower than a preset shadow threshold, the control module sends a enhancement command. The light source increases the brightness by increasing the driving current and adjusts the illumination angle of the light source to make the light cover the shadow area more evenly. After each brightness adjustment, the area array image acquisition component acquires local test images in real time. By analyzing the uniformity of grayscale distribution in each sub-region of the image, if the uniformity does not meet the preset standard, the adjustment process is repeated until the light in each area of ​​the ground is consistent, ensuring clear imaging of planar traces.

[0043] Specifically, the planar information acquisition module performs edge enhancement processing on the acquired planar trace image based on an image preprocessing mechanism. It adopts a direction enhancement algorithm based on gray-level gradient, which strengthens the gray-level difference of edge pixels by identifying the gray-level change trend of the trace edge. By extracting the main color information of the ground background, a background color template is constructed. Noise pixels in the planar trace image that deviate from the background color template but do not belong to the trace features are replaced with the gray-level values ​​of adjacent background pixels, thus preserving the color features of the trace itself. The preprocessed image is quality evaluated by calculating the contrast value between the trace area and the background area. If the contrast value reaches a preset threshold, the preprocessed image is marked and stored.

[0044] The first step involves edge enhancement processing of the acquired planar trace images. A gray-level gradient-based directional enhancement algorithm is used to identify the gray-level variation trends at the trace edges (such as the direction of footprint patterns and the extension direction of scratches), specifically enhancing the gray-level difference of edge pixels to improve the distinction between the trace edges and the background. The second step involves targeted noise removal. By extracting the dominant color information of the ground background, a background color template is constructed. Noise pixels in the image that deviate significantly from the background color template but are not trace features (such as ground stains or dust reflections) are replaced with the gray-level values ​​of adjacent background pixels, while preserving the color characteristics of the trace itself (such as the dark red of bloodstains or the transparent color shift of grease). The third step involves quality assessment of the preprocessed image. The contrast value between the trace area and the background area is calculated. If the contrast is lower than a preset threshold, the edge enhancement and noise removal steps are automatically re-executed. If the contrast meets the standard, the preprocessed image is marked as "to be analyzed" and stored to ensure clear extraction of trace details during subsequent evidence identification.

[0045] Specifically, during the process of the multi-band special light source component stimulating the trace reaction, a first-band light source is preset according to the matching relationship between the material of the facade and the type of trace; after the first-band light source is activated, the linear array scanning acquisition component scans the target facade area line by line and transmits the scanned image in real time; by calculating the grayscale variance of the image, it is determined that no potential trace is detected, and the system switches to the second-band light source and adjusts the scanning parameters of the linear array scanning acquisition component.

[0046] The control module, based on the matching relationship between the facade material (e.g., wall paint, metal surface, wood paneling) and common trace types (e.g., bloodstains, grease, fibers, ink), presets a first-band light source (e.g., for bloodstains on painted walls, the ultraviolet light band is prioritized). After activating the first-band light source, the linear array scanning acquisition component scans the target facade area line by line, simultaneously transmitting the scanned image to the trace detection unit of the control module in real time. The trace detection unit calculates the grayscale variance of the image (if the grayscale variance is lower than a preset threshold, it indicates that there are no obvious trace features in the area), and if... If no potential traces are detected, the control module automatically triggers the band switching mechanism of the multi-band special light source component, switching to the second band light source (such as the laser band). At the same time, it adjusts the scanning parameters of the linear array scanning acquisition component (such as scanning step size and exposure time) to adapt to the light intensity characteristics of the new band. The scanning and detection process is repeated until traces are detected (grayscale variance meets the standard) or all preset bands are scanned. If no traces are detected in the full band scan, the control module automatically records the location information of the facade area and marks it as "awaiting manual verification" for subsequent on-site inspection personnel to focus on.

[0047] Specifically, after capturing physical evidence, the planar information acquisition module or the facade information acquisition module sends a trigger signal including the three-dimensional coordinates of the physical evidence and the type of physical evidence; after receiving the signal, it analyzes the coordinates of the physical evidence, sends a command to the multi-lens circular array panoramic camera, and adjusts the lens in the corresponding direction to align with the local area where the physical evidence is located; it adjusts the acquisition parameters of the panoramic camera, and after the acquisition is completed, it calculates the sharpness value of the local image.

[0048] After capturing physical evidence, the planar information acquisition module or the elevation information acquisition module immediately sends a trigger signal to the control module containing the three-dimensional coordinates of the evidence (calculated by combining the positioning information of the intelligent mobile platform with the relative position of the acquisition module) and the type of evidence (such as footprints or tool marks). Upon receiving the signal, the control module analyzes the coordinates of the evidence and sends a command to the lens turning mechanism of the multi-lens circular array panoramic camera to adjust the lens in the corresponding direction to align with the local area where the evidence is located. At the same time, the control module adjusts the acquisition parameters of the panoramic camera: on the one hand, it increases the acquisition frequency of the local area (e.g., once per second for regular areas and 3-5 times per second for key areas), acquiring multiple frames of images through multiple acquisitions to facilitate subsequent selection of the clearest frame; on the other hand, it extends the single acquisition time to improve the signal-to-noise ratio of the image by increasing the lens exposure time and reducing ambient light interference. After the acquisition is completed, the control module evaluates the sharpness of the multiple frames of local images (by calculating the image sharpness value), selects the image with the highest sharpness value as the key acquisition result, stores it together with the regular panoramic images, and marks the location of the evidence in the key images, providing high-definition local data support for subsequent panoramic stitching and 3D modeling.

[0049] Specifically, the image stitching processing mechanism extracts the overlapping area of ​​the lens images, selects color feature points within the overlapping area, calculates the color parameters of the feature points in the lens images using a color analysis algorithm, generates a color adjustment mapping table based on the color parameters of one lens, performs color correction on other lens images according to the color adjustment mapping table, and calculates the color difference between pixels on both sides of the stitching line.

[0050] The first step involves the image stitching processing unit extracting the overlapping areas (i.e., the intersections of adjacent image frames) from the images captured by each lens, and selecting multiple color feature points within these overlapping areas (such as object edges with obvious color transitions or background areas with fixed hues). The second step uses a color analysis algorithm to calculate the color parameters (including saturation, brightness, and hue) of the feature points in each image frame. Using the color parameters of one lens (usually the middle lens, with the optical axis closest to horizontal) as a benchmark, a color adjustment mapping table is generated. The third step involves color correction of the images from other lenses based on the mapping table: for areas with saturation deviations, the gain value of the color channels is adjusted to achieve uniformity; for areas with brightness deviations, the pixel grayscale range is adjusted using a grayscale stretching algorithm; for areas with hue deviations, color wheel rotation correction is used to achieve consistent hue. After color calibration, the image stitching processing unit checks the pixel transition at the stitching point (calculating the color difference between pixels on both sides of the stitching line). If the difference exceeds a preset threshold (indicating a color banding), feature points are reselected and the mapping table is adjusted until the color transition at the stitching point is natural, resulting in a distortion-free panoramic image with consistent overall color.

[0051] Specifically, the spatial coordinate association component binds the panoramic image with the on-site three-dimensional spatial location, obtains the coordinate data of the current acquisition location, extracts the three-dimensional coordinates of fixed feature points on-site from the laser point cloud data, calculates the difference between the coordinate data and the three-dimensional coordinates, and calculates the deviation compensation value using a weighted average method; adjusts the coordinate binding parameters of the panoramic image according to the deviation compensation value, and corrects the coordinates of each pixel in the panoramic image to the original bound coordinates and the deviation compensation value.

[0052] The position deviation correction steps include: First, the spatial coordinate association component obtains the coordinate data of the current acquisition position from the positioning module of the intelligent mobile platform, and simultaneously extracts the three-dimensional coordinates of multiple fixed feature points (such as corners, ground protrusions, and edges of fixed equipment) from the laser point cloud data; Second, it calculates the difference (i.e., the deviation value) between the coordinates provided by the positioning module and the coordinates of the feature points in the laser point cloud, and uses a weighted average method to calculate the deviation compensation value (giving higher weights to feature points closer to the intelligent mobile platform, as the coordinates of nearby feature points are more accurate); Third, it adjusts the coordinate binding parameters of the panoramic image according to the deviation compensation value, correcting the coordinates of each pixel in the panoramic image to "original bound coordinates + deviation compensation value"; After the correction is completed, the spatial coordinate association component re-compares the corrected panoramic image coordinates with the laser point cloud feature point coordinates. If the coordinate differences of all feature points are within the allowable error range (i.e., the deviation correction is effective), the binding relationship is confirmed; If there are still cases where the error exceeds the limit, the feature points are automatically re-extracted and the deviation calculation and correction process is repeated to ensure the accuracy of the binding between the panoramic image and the three-dimensional spatial position of the site, thereby ensuring the reliability of the spatial position information of the physical evidence.

[0053] Specifically, the process of supplementing light in the non-visible light band by the darkroom information acquisition module includes: scanning the darkroom environment to generate a global grayscale image, generating grayscale abnormal areas, and marking the boundary coordinates of suspected traces; sending a supplementing light adjustment command to the darkroom information acquisition module, wherein the supplementing light range in the non-visible light band covers the area enclosed by the boundary coordinates of the suspected traces.

[0054] The low-light enhancement image sensor performs a preliminary scan of the darkroom environment, generating a global grayscale image. The trace recognition unit of the control module analyzes grayscale anomaly areas in the grayscale image (such as suspected trace areas with grayscale values ​​higher than the surrounding background) and marks the boundary coordinates of the suspected traces. In the second step, the control module sends a command to the supplementary lighting adjustment mechanism of the darkroom information acquisition module. The supplementary lighting adjustment mechanism adjusts the angle and aperture of the supplementary lighting lens to ensure that the supplementary lighting range in the non-visible light band accurately covers the area enclosed by the boundary coordinates of the suspected traces (the supplementary lighting range is slightly larger than the trace area to avoid edge omissions). (1) Light); The third step is that after the supplementary lighting is started, the low-light enhancement image sensor acquires local images after supplementary lighting in real time. The control module analyzes the illumination uniformity of the trace area in the image (if there are local areas that are too bright or too dark, it means that the supplementary lighting range is not fully adapted). If the uniformity is not up to standard, the supplementary lighting adjustment mechanism readjusts the lens angle and aperture until the trace area is illuminated evenly. At the same time, the control module ensures that the supplementary lighting range does not exceed the suspected trace area by too much, so as to avoid the supplementary lighting covering other potential trace areas, which would cause the area to be unidentifiable due to supplementary lighting interference, thus ensuring the simultaneous detection of multiple traces in the darkroom environment.

[0055] Specifically, the noise suppression processing component reduces grain noise in night view images by comparing the night view images before and after noise suppression to locate the trace detail areas lost due to noise suppression; the detail restoration algorithm constructs a grayscale distribution model of the trace details based on the grayscale change trend of adjacent pixels around the lost area, and calculates the pixel grayscale values ​​of the lost area through interpolation; the restored image is compared with the original image before noise suppression, and if the deviation is less than a preset value, the restored image is stored.

[0056] The image detail restoration process includes the following steps: First, the noise suppression processing component transmits the night view images before and after noise suppression to the detail comparison unit of the control module. The detail comparison unit locates the trace detail areas lost due to noise suppression (such as fine lines on the edge of footprints, changes in the depth of tool marks, and areas where pixel grayscale values ​​change abruptly) through pixel-level comparison. Second, the detail restoration algorithm constructs a grayscale distribution model of the trace details based on the grayscale change trend of adjacent pixels around the lost area (such as the grayscale gradient direction and grayscale difference of adjacent pixels). It then uses interpolation to calculate and supplement the pixel grayscale values ​​of the lost area, ensuring that the restored details transition naturally with the surrounding pixels. Third, after the detail restoration is completed, the control module performs feature comparison between the restored image and the original image before noise suppression (such as extracting feature parameters such as the contour length and number of lines of the trace). If the deviation between the restored feature parameters and the feature parameters of the original image is less than a preset value (indicating accurate detail restoration), the restored image is stored. If the deviation is large, the parameters of the detail restoration algorithm are readjusted, and the restoration process is executed again to ensure the integrity of the details of the evidence image in the darkroom scene.

[0057] Specifically, after the planar information acquisition module captures two-dimensional planar traces on the ground, it generates a planar trace image and sends a linkage trigger signal, including the trace acquisition timestamp, ground two-dimensional coordinates, and preliminary trace classification results; the two-dimensional coordinates of the trace are converted into spatial three-dimensional coordinates and sent to the spatial information acquisition module; after acquiring a panoramic image of the trace area, the planar trace image is associated and stored with the corresponding spatial panoramic image: the same association identifier code is assigned to both, the acquisition timestamp, acquisition location coordinates, and acquisition parameters are recorded, and an association index table is established.

[0058] Specifically, when the facade information acquisition module scans the facade in a low-light environment where the ambient light intensity is lower than a preset low-light threshold, it uses an image edge detection algorithm to identify suspected trace areas with abrupt changes in grayscale values ​​and discontinuous contours. When the image sharpness value is lower than a preset standard, it sends a linkage request signal, which includes the facade location coordinates and imaging parameters of the suspected trace area. The darkroom information acquisition module is then activated to perform secondary acquisition on the suspected trace area. After generating a night view image, the suspected trace image undergoes feature comparison.

[0059] After constructing a three-dimensional model of the site, the spatial information acquisition module annotates the airborne harmful gas data as semantic information to generate a gas concentration heat map; it marks the physical evidence locations in the three-dimensional model of the site, associates the corresponding planar acquisition images, elevation acquisition images, and darkroom acquisition images with the physical evidence, and generates a physical evidence association archive including multi-dimensional acquisition data.

[0060] The intelligent mobile platform also integrates a hazardous gas detection module. This module starts synchronously with the spatial information acquisition module, collecting real-time data on hazardous gases in the air (including gas type, concentration value, and acquisition location coordinates). The gas type detection covers common hazardous gases found in criminal cases (such as carbon monoxide, hydrogen sulfide, and volatile organic compounds). The acquisition location coordinates are obtained through the intelligent mobile platform's positioning accompanying module and are consistent with the coordinate system of the spatial three-dimensional model. During the acquisition process, if the concentration of a gas exceeds a preset safety threshold (set according to the "Indoor Air Quality Standard" GB / T 18883-2022), the hazardous gas detection module automatically increases the sampling frequency for that area and simultaneously sends an abnormal concentration warning signal to the control module. After acquisition, the gas data is validated, and data exceeding the detection range or with acquisition coordinate deviations greater than the spatial model coordinate deviation threshold are discarded, resulting in a qualified gas dataset G={(G1, T1, P1), (G2, T2, P2), ..., (G n T n P n )}(G n T represents the type and concentration of the gas. nTo collect timestamps, P n (To collect the three-dimensional coordinates of the location).

[0061] Semantic annotation and heatmap generation of hazardous gas data:

[0062] The first step is to semantically annotate the qualified gas dataset G: Divide the data into concentration ranges based on gas concentration values ​​(e.g., low concentration: 0~C1, medium concentration: C1~C2, high concentration: C2~C_max, where C1 and C2 are set separately according to the gas type, e.g., carbon monoxide C1=10mg / m³, C2=30mg / m³), and assign these ranges to each data point G. n The first step is to attach semantic tags (e.g., "carbon monoxide - low concentration", "volatile organic compounds - high concentration"); the second step is to generate a gas concentration heat map: call the image rendering unit of the spatial information acquisition module, and use the spatial coordinates of the on-site 3D model as a reference to plot the labeled gas data points P. n The gas concentration thermal layer is generated by mapping the corresponding location to the 3D model and using color gradient rendering (e.g., green for low concentration, yellow for medium concentration, and red for high concentration). The specific concentration value and collection timestamp are marked in each concentration anomaly area (concentration ≥ C2) on the thermal map. The third step is to bind the thermal map to the 3D model: through the spatial coordinate association component, the gas concentration thermal layer is texture-mapped to the on-site 3D model to ensure that the thermal map is updated synchronously with the scaling and rotation of the 3D model, realizing the visual association of "spatial location - gas concentration - semantic label".

[0063] In this embodiment, for the data collected from the four modules of plan view, elevation view, darkroom view, and space view, a "parametric threshold control + automated algorithm processing" approach is adopted to ensure that the data quality meets the standards. The core parameters are defined as follows:

[0064] Set the preprocessing quality assessment thresholds as follows: the contrast threshold for planar / elevation images is T1 (if it is below this value, it needs to be reprocessed), the signal-to-noise ratio threshold for darkroom images is T2 (if it is below this value, noise suppression needs to be added), and the coordinate deviation threshold for spatial models is T3 (if it exceeds this value, the binding relationship needs to be corrected).

[0065] Let the algorithm control parameters be: the gradient coefficient of the edge enhancement algorithm be K1 (controlling the edge enhancement intensity), the background matching threshold of the noise removal algorithm be K2 (above this value is judged as background pixels), and the interpolation step size of the detail restoration algorithm be Δx (Δx=1px, 1px represents the physical distance of 1 pixel, ensuring natural pixel transition).

[0066] Planar module data preprocessing

[0067] Edge enhancement processing: Invoke the orientation enhancement algorithm based on grayscale gradient, and input the pixel matrix M of the footprint image A. a(with a resolution of W×H), calculate the grayscale gradient G of each pixel a G(i, j) = |M a (i + 1, j) - M a (i, j)| + |M a (i, j + 1) - M a (i, j)| (where i ∈ [1, W], j ∈ [1, H]);

[0068] For pixels where G a (i, j) is less than K1 × average gradient value, according to the formula M a ’(i, j) = M a (i, j) + K1 × (average gradient value - G a (i, j)), enhance the grayscale difference value to obtain the enhanced image A1.

[0069] Noise removal processing: Extract the RGB average values (R0, G0, B0) of the main color of the ground background to construct the background color template T0; for each pixel (R(i, j), G(i, j), B(i, j)) in A1, calculate the matching degree S(i, j) = √[(R(i, j) - R0)² + (G(i, j) - G0)² + (B(i, j) - B0)²] / 255;

[0070] If S(i, j) ≥ K2, it is determined as a background noise pixel, and its grayscale value is replaced with the average value of the background pixels in the adjacent 3×3 area to obtain the denoised image A2.

[0071] Quality verification: Calculate the contrast C of A2 a = (maximum grayscale value of the image - minimum grayscale value) / (maximum grayscale value of the image + minimum grayscale value). If C a ≥ T1, it is determined that the preprocessing is qualified; if C a < T1, automatically adjust K1 to 1.5, repeat the steps until C a ≥ T1, and finally output the qualified planar data.

[0072] Preprocessing of elevation module data

[0073] Tilt correction processing: Due to the tilt of the wall, there is a perspective distortion in the trace image B. Obtain the wall tilt angle α (the angle between image B and the horizontal direction) through the angle sensor of the elevation module. According to the perspective transformation formula:

[0074] [x’, y’]ᵀ = [cosα -sinα; sinα cosα] × [x, y]ᵀ (x, y are the original pixel coordinates of B, x’, y’ are the corrected coordinates) to obtain the corrected image B1.

[0075] Color removal and contrast optimization: In-plane module steps, input the pixel matrix Mᵦ of B1, calculate the image B2 after color removal, verify that the contrast Cᵦ ≥ T1, and output the qualified facade data.

[0076] Data preprocessing of the darkroom module

[0077] Noise suppression processing: Input the grayscale matrix of the initial darkroom image C , and use the Gaussian filtering algorithm to calculate according to the formula . G(Δi,Δj) is the weight value of the 5×5 Gaussian kernel function at the offset (Δi,Δj), and Δi and Δj are the offsets relative to the central pixel (i,j), and their value ranges are both in the integer interval [-2,2]. is the grayscale value of a specific adjacent pixel within the 5x5 neighborhood centered on the pixel (i,j). Remove high-frequency granular noise to obtain the denoised image C1.

[0078] Detail restoration processing: Compare the pixel differences between C1 and the original image C to locate the missing detail area D (manifested as a cluster of pixels with sudden changes in grayscale values in C1);

[0079] For each pixel (i, j) in D, construct a grayscale distribution model according to the grayscale trends of the adjacent 8 pixels: ; where, Δi and Δj are the offsets relative to the central pixel (i,j), and their value range is in the integer interval [-1,1], but the case of (Δi,Δj)=0 (i.e., the central pixel itself) needs to be excluded, and interpolation is supplemented to obtain the restored image C2.

[0080] Quality verification: Calculate the signal-to-noise ratio SNR = 10 lg(signal power / noise power) of C2. If SNR ≥ T2, output the qualified darkroom data; if SNR < T2, increase the Gaussian kernel size to 7×7 and repeat the steps.

[0081] Data preprocessing of the spatial module

[0082] Coordinate deviation correction: Input the coordinate set P = {(X1, Y1, Z1), (X2, Y2, Z2),..., (X n , Y n , Z n )} (n is the number of model vertices) generated by the spatial module, and extract 3 fixed feature points Q1, Q2, Q3 from the laser point cloud data (such as the corner of the wall, the raised object on the ground, and the known coordinates are (Q 1x , Q1ᵧ, Q 1z ), (Q 2x , Q2ᵧ, Q 2z ), (Q 3x , Q3ᵧ, Q 3z ));

[0083] Calculate the deviation values of the corresponding feature points P1, P2, P3 and Q1, Q2, Q3 in the calculation model , and use the weighted average method to calculate the deviation compensation value Δ=(w1ΔP1+w2ΔP2+w3ΔP3) / 3. Here, since the weight w k =1 / ΔP k (k = 1, 2, 3) has been normalized, and the denominator is the sum of the weights, that is, w1+w, so the formula can be simplified to the shown form.

[0084] According to the formula Correct all vertex coordinates (Δx, , Δz is the component of Δ on the X / Y / Z axis), and obtain the corrected model.

[0085] Quality verification: Recalculate the deviation ΔP' of the corrected feature points P'1, P'2, P'3 and Q1, Q2, Q3 k , if all ΔP' k ≤T3, determine that the model is qualified; if there exists ΔP' k >T3, re-extract the feature points Q4, Q5, and repeat the deviation calculation and correction steps.

[0086] The above is only a preferred embodiment of the present invention, and it does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the disclosed technical content within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An autonomous vehicle-based evidence investigation system based on panoramic image reconstruction and trace acquisition, characterized in that, It includes an intelligent mobile platform, a planar information acquisition module, a facade information acquisition module, a spatial information acquisition module, and a darkroom information acquisition module; The intelligent mobile platform is a mobile platform based on complex environments, carrying an information collection module; The planar information acquisition module captures two-dimensional planar traces on the ground through the area array image acquisition component, eliminates ground shadow interference through diffused light source output, and the automatic focus adjustment component adjusts the acquisition focal length according to the ground distance to generate acquisition and outline preservation images of planar physical evidence. The facade information acquisition module is based on a linear array scanning acquisition component. It scans the vertical or inclined surface of the facade line by line. A multi-band special light source component excites the fluorescent reaction or contrast imaging of the traces. An angle adjustment mechanism drives the linear array scanning acquisition component to rotate within a preset range to perform trace search and extraction. The spatial information acquisition module acquires airborne harmful gas data and generates a gas concentration heat map. The circumferentially distributed lenses simultaneously acquire on-site panoramic images. The on-site panoramic images are seamlessly fused through an image stitching processing mechanism to generate a distortion-free panoramic image. The spatial coordinate association component combines the positioning information of the intelligent mobile platform to bind the panoramic image with the three-dimensional spatial location of the site. Based on laser point cloud data, a texture mapping between the panoramic image and the point cloud model is generated to construct a three-dimensional model of the site that includes the spatial location information of the physical evidence. In a darkroom environment, the darkroom information acquisition module converts weak light signals into night vision images, supplements the light with non-visible light bands, and reduces the grain noise of the night vision images by the noise suppression processing component, thereby generating images and records of footprints, tool marks, and trace evidence in the darkroom scene.

2. The system according to claim 1, characterized in that, The area array image acquisition component divides the acquisition area into equal detection sub-regions and samples the ground reflected light intensity of the detection sub-regions; if the reflected light intensity of the detection sub-region exceeds a preset reflection threshold, a reduction command is sent to the diffuse light source; the area array image acquisition component acquires local test images in real time and generates the grayscale distribution uniformity of each sub-region in the image.

3. The system according to claim 1, characterized in that, The planar information acquisition module performs edge enhancement processing on the acquired planar trace image based on the image preprocessing mechanism. It adopts a direction enhancement algorithm based on gray-level gradient, which strengthens the gray-level difference of edge pixels by identifying the gray-level change trend of the trace edge. By extracting the main color information of the ground background, a background color template is constructed. The noisy pixels in the planar trace image that deviate from the background color template but do not belong to the trace features are replaced with the gray-level values ​​of adjacent background pixels, thus preserving the color features of the trace itself. The quality of the preprocessed image is evaluated by calculating the contrast value between the trace area and the background area. If the contrast value reaches a preset threshold, the preprocessed image is marked and stored.

4. The system according to claim 1, characterized in that, During the process of the multi-band special light source component stimulating the trace reaction, a first-band light source is preset according to the matching relationship between the material of the facade and the type of trace. After the first-band light source is activated, the linear array scanning acquisition component scans the target facade area line by line and transmits the scanned image in real time. By calculating the grayscale variance of the image, it is determined that no potential trace is detected, and the system switches to the second-band light source and adjusts the scanning parameters of the linear array scanning acquisition component.

5. The system according to claim 1, characterized in that, After capturing physical evidence, the planar information acquisition module or the facade information acquisition module sends a trigger signal including the three-dimensional coordinates of the physical evidence and the type of physical evidence. After receiving the signal, it analyzes the coordinates of the physical evidence, sends a command to the multi-lens circular array panoramic camera, and adjusts the lens in the corresponding direction to align with the local area where the physical evidence is located. It adjusts the acquisition parameters of the panoramic camera, and after the acquisition is completed, it calculates the sharpness value of the local image.

6. The system according to claim 1, characterized in that, The image stitching processing mechanism extracts the overlapping area of ​​the lens images, selects color feature points within the overlapping area, calculates the color parameters of the feature points in the lens images using a color analysis algorithm, generates a color adjustment mapping table based on the color parameters of one lens, and performs color correction on other lens images according to the color adjustment mapping table, calculating the color difference between pixels on both sides of the stitching line.

7. The system according to claim 1, characterized in that, The spatial coordinate association component binds the panoramic image to the on-site three-dimensional spatial location, obtains the coordinate data of the current acquisition location, extracts the three-dimensional coordinates of fixed feature points on-site from the laser point cloud data, calculates the difference between the coordinate data and the three-dimensional coordinates, and calculates the deviation compensation value using a weighted average method. Based on the deviation compensation value, the coordinate binding parameters of the panoramic image are adjusted to correct the coordinates of each pixel in the panoramic image to the original bound coordinates and the deviation compensation value.

8. The system according to claim 1, characterized in that, The process of supplementing light in the non-visible light band by the darkroom information acquisition module includes: scanning the darkroom environment, generating a global grayscale image, marking grayscale abnormal areas, and recording the boundary coordinates of suspected traces; sending a supplementing light adjustment command to the darkroom information acquisition module, wherein the supplementing light range in the non-visible light band covers the area enclosed by the boundary coordinates of the suspected traces.

9. The system according to claim 1, characterized in that, The noise suppression processing component reduces grain noise in night view images by: comparing night view images before and after noise suppression to locate trace detail areas lost due to noise suppression; the detail restoration algorithm constructs a grayscale distribution model of trace details based on the grayscale change trend of adjacent pixels around the lost area, and calculates the pixel grayscale values ​​of the lost area through interpolation; the restored image is compared with the original image before noise suppression, and if the deviation is less than a preset value, the restored image is stored.

10. The system according to claim 1, characterized in that, After capturing two-dimensional planar traces on the ground, the planar trace image is generated and a linkage trigger signal is sent, including the trace acquisition timestamp, ground two-dimensional coordinates, and preliminary trace classification results. The two-dimensional coordinates of the traces are converted into three-dimensional spatial coordinates and sent to the spatial information acquisition module. After acquiring a panoramic image of the trace area, the planar trace image is associated and stored with the corresponding panoramic image: the same association identifier code is assigned to both, the acquisition timestamp, acquisition location coordinates, and acquisition parameters are recorded, and an association index table is established.

11. The system according to claim 1, characterized in that, When the facade information acquisition module scans the facade in a low-light environment where the ambient light intensity is lower than a preset low-light threshold, it uses an image edge detection algorithm to identify suspected trace areas with abrupt changes in grayscale values ​​and discontinuous contours. When the image sharpness value is lower than a preset standard, it sends a linkage request signal, including the facade location coordinates and imaging parameters of the suspected trace area. The darkroom information acquisition module is then activated to perform secondary acquisition on the suspected trace area. After generating a night view image, the suspected trace image is subjected to feature comparison.

12. The system according to claim 1, characterized in that, After the spatial information acquisition module constructs a three-dimensional model of the site, it annotates the air pollutant data as semantic information to generate a gas concentration heat map. The locations of physical evidence in the three-dimensional model of the scene are marked, and the corresponding planar images, elevation images, and darkroom images of the physical evidence are associated to generate a physical evidence association archive that includes multi-dimensional acquisition data.

Images