A human specimen transfer vehicle following method based on camera vision

Abstract

The present application belongs to the technical field of image processing, and relates to a human specimen transfer trolley following method based on camera vision. The method collects an environment color image and converts it into a gray image, calculates a light spot confidence based on the brightness characteristics and the gradient distribution characteristics of the gradient image, and then generates a binary mask. Image feature points are extracted in the unobstructed area, and abnormal feature points are effectively removed by using the direction consistency of the optical flow vector and the perspective projection law. Then, the physical Euclidean distance is weighted processed according to the number of effective feature points and the average optical flow re-projection error, and an effective following distance is calculated. Finally, the real-time load mass is obtained by a pressure sensor to calculate the variable mass safety braking distance, and an exponential flexible speed command is generated by combining the difference of the effective following distance. The present application realizes the smooth start and stop of the transfer trolley under the heavy liquid transportation working condition, eliminates the start and stop jerk and the hidden danger of anticorrosive liquid splashing, and improves the operation safety.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a method for following a human specimen transposition vehicle based on camera vision. Background Technology

[0002] In modern medical education and research institutions, the transport of human specimens and associated preservative storage facilities is a demanding and risky task. To meet stringent hygiene and corrosion prevention standards, the main structure of these transport trolleys is generally made of stainless steel and requires a high rated load-bearing capacity. With the continuous evolution of medical automation technology, mobile robots with automatic following capabilities are gradually being introduced into this field. Current automatic following technology primarily relies on depth cameras to acquire images, combined with cutting-edge deep learning neural network algorithms for human target feature detection, and then uses traditional PID algorithms to adjust the motor speed based on the target distance deviation.

[0003] However, the actual scenarios of medical teaching and specimen storage are extremely complex, posing a severe challenge to existing visual perception technologies. Inside dissection rooms and specimen storage rooms, numerous stainless steel refrigerated cabinets, dissection tables, and various medical instrument carts are densely packed. Under strong indoor lighting or professional shadowless lamps, these stainless steel surfaces exhibit strong specular reflections. Existing visual processing algorithms are prone to misinterpreting these bright reflective points as spatial information with physical depth, or incorrectly identifying the clear reflection of the operator projected onto the stainless steel surface as the actual target to be followed. This can lead to frequent malfunctions of the transfer cart, and even dangerous situations where it deviates from its intended path and collides with surrounding expensive equipment.

[0004] Besides the complex visual interference environment, the dynamic physical characteristics of the transposition vehicle during actual operation also constrain the stability of the following system. The overall mass of a fully loaded human specimen transposition vehicle is typically enormous, often far exceeding that of an unloaded vehicle, and the items it carries contain large amounts of liquid chemical preservatives. Existing linear following control algorithms fail to adequately consider the physical impact of significant changes in vehicle load on braking distance, resulting in sluggish power response during start-up and, due to the enormous inertia caused by the heavy load, inability to accurately stop at the designated safe position during parking, even frequently exhibiting abrupt stops and jerks. Such kinetic impacts can easily cause specimens to slip or toxic preservative chemicals to splash out, leading to serious safety accidents. Summary of the Invention

[0005] The purpose of this invention is to propose a camera vision-based method for following a human specimen transfer vehicle, in order to solve the technical problems of visual recognition failure of existing human specimen transfer vehicles in the high reflectivity of stainless steel environments and the easy occurrence of start-stop jerking and liquid splashing during heavy-duty liquid transportation.

[0006] The technical solution of the camera vision-based human specimen transposition vehicle following method provided by this invention is as follows:

[0007] Camera vision-based methods for following human specimen transposition vehicles include:

[0008] The environmental image in front of the human specimen transfer vehicle is acquired, the color image in the environmental image is converted into a grayscale image and the gradient map is calculated. The confidence of the light spot is calculated based on the brightness characteristics of the pixels in the grayscale image and the gradient distribution characteristics in the gradient map. A binary mask is generated based on the confidence of the light spot to shield the specular reflection area.

[0009] Image feature points are extracted in areas not obscured by binarization masks. The optical flow vectors of the image feature points are calculated between consecutive frames. Based on the consistency of the direction of the optical flow vectors and the perspective projection rules, real targets and false reflections are distinguished. Abnormal feature points are removed to lock onto the following target.

[0010] Based on the number of valid feature points retained in the current frame and the average optical flow reprojection error of the valid feature points, the measured physical Euclidean distance is weighted to calculate the effective following distance after visual confidence weighting.

[0011] The real-time load mass is obtained by the pressure sensor in the hydraulic lifting circuit of the human specimen transfer vehicle, and the variable mass safe braking distance is calculated by combining it with the current driving speed of the human specimen transfer vehicle; an exponential flexible speed command is generated based on the difference between the effective following distance and the variable mass safe braking distance to control the human specimen transfer vehicle to follow the target.

[0012] Beneficial effects: This invention generates a binary mask by calculating the confidence level of the light spot, which can effectively shield the interference of the specular reflection area in the stainless steel environment; it eliminates false reflections by utilizing the directional consistency of the optical flow vector and the perspective projection law, ensuring the uniqueness of target locking; it performs weighted processing of the physical Euclidean distance based on the number of effective feature points and the average optical flow reprojection error, preventing vehicle malfunctions caused by sudden changes in visual signals; it obtains real-time load mass through a pressure sensor to calculate the variable mass safe braking distance and generates an exponential flexible speed command, enabling the vehicle to perceive its own inertia, thereby solving the problems of start-stop jerking and liquid splashing caused by dynamic mismatch under heavy-duty liquid transportation conditions.

[0013] Furthermore, the formula for calculating the effective following distance is:

[0014]

[0015] In the formula, This represents the effective following distance after weighting by the visual confidence level; Represents the physical Euclidean distance; This indicates the number of valid feature points retained in the current frame; Indicates the saturation constant of the characteristic quantity; This represents the average optical flow reprojection error of the effective feature points; This represents the error penalty coefficient.

[0016] Beneficial effects: This formula introduces two key variables—the number of effective feature points and the average optical flow reprojection error—which directly convert image processing quality indicators into distance control signals. When environmental interference reduces the number of effective feature points or increases the tracking and matching error, the calculated effective following distance value will automatically decrease. This fault-oriented safety logic can promptly trigger deceleration, preventing the risk of vehicle loss of control due to deterioration in vision system signal quality.

[0017] Furthermore, the formula for calculating the variable mass safety braking distance is as follows:

[0018]

[0019] In the formula, This indicates the variable mass safety braking distance; This represents the total vehicle mass calculated based on the real-time load mass. Indicates the current speed of the human specimen transfer vehicle; Indicates the kinetic energy braking coefficient; Indicates the system response time constant; This indicates a rigid safety margin.

[0020] Beneficial effects: Through this solution, the control system can sense the real-time total weight and current speed of the human specimen transfer vehicle, and dynamically adjust the vehicle's safe braking distance. This ensures that the deceleration command can be executed in advance under extreme conditions of being fully loaded with antiseptic liquid and specimens, avoiding the collision hazards caused by insufficient braking distance in traditional fixed parameter control.

[0021] Furthermore, the generation of the exponential flexible speed command based on the difference between the effective following distance and the variable mass safe braking distance satisfies the following formula:

[0022]

[0023] In the formula, This refers to the target speed command output to the motor controller in the exponential flexible speed command. This indicates the preset maximum cruising speed; This represents the flexible damping length constant; It is the base of the natural logarithm;

[0024] When the effective following distance is less than or equal to the variable mass safety braking distance, the value of the target speed command is 0.

[0025] Beneficial effects: This formula uses the smoothing characteristics of exponential functions to replace traditional linear control. When the effective following distance approaches the variable mass safety braking distance, the target speed command smoothly converges to 0. This control logic avoids the sudden acceleration phenomenon of the human specimen transposition vehicle during operation, prevents the loaded anti-corrosion chemical liquid from splashing out due to mechanical shaking, and ensures the level of biosafety protection.

[0026] Further, the step of calculating the confidence level of the light spot based on the brightness characteristics of pixels in the grayscale image and the gradient distribution characteristics in the gradient map includes:

[0027] Statistically analyze the grayscale value and local gradient magnitude of each pixel in the grayscale image;

[0028] When the gray value of the pixel is greater than a preset gray value threshold and the local gradient magnitude is greater than a preset gradient threshold, the pixel is determined to be a speckle interference point, and the area containing the speckle interference point is determined to be a speckle confidence area.

[0029] Furthermore, the extraction of image feature points employs the AKAZE algorithm, utilizing nonlinear diffusion filtering to construct a scale space.

[0030] Beneficial effect: Using nonlinear diffusion filtering to construct scale space to extract image feature points can better preserve the texture edge details of metal surfaces.

[0031] Further, the removal of abnormal feature points includes: calculating the mean vector of the optical flow vectors of all extracted image feature points; calculating the Euclidean distance between the optical flow vector of each image feature point and the mean vector; and identifying image feature points whose Euclidean distance is greater than a preset statistical threshold as abnormal feature points and removing them.

[0032] Furthermore, the step of obtaining the real-time load mass through the pressure sensor in the hydraulic lifting circuit of the human specimen transfer vehicle includes: reading the fluid pressure value output by the pressure sensor installed at the bottom of the hydraulic cylinder on the human specimen transfer vehicle, and converting the fluid pressure value into the real-time load mass according to Pascal's principle.

[0033] Furthermore, the total mass of the vehicle includes the vehicle's own weight and the mass of the specimen and liquid calculated from the readings of the pressure sensor;

[0034] The kinetic braking coefficient is a constant set based on the energy dissipation capacity of the hydraulic braking system configured on the human specimen transfer vehicle.

[0035] Further, the calculation of the gradient map includes: using the Sobel edge detection operator to calculate the spatial derivatives of the grayscale image in the horizontal and vertical directions respectively; obtaining the sum of squares of the spatial derivatives in the horizontal and vertical directions, and performing a square root operation on the sum of squares to synthesize the gradient map.

[0036] The beneficial effects of this invention are as follows: This invention can eliminate the influence of specular reflection areas and high-light interference points on visual confidence, accurately eliminate false reflections and background noise, and achieve stable locking of the target being followed; at the same time, it can endow the human specimen transfer vehicle with the ability to sense real-time load mass changes, and maintain consistent target speed command response and smooth convergence trend under different load conditions including only the vehicle body weight or the addition of specimen and liquid mass, effectively eliminating the start-stop jerking feeling of hydraulic drive, and ensuring the stability and safety of the human specimen transfer vehicle in the process of following the target. Attached Figure Description

[0037] Figure 1 This is a schematic flowchart of the human specimen transposition vehicle following method based on camera vision of the present invention;

[0038] Figure 2 This is a schematic diagram of the consistency analysis of optical flow vectors of visual feature points in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the safety braking domain mapping under variable load conditions in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram comparing the following response and smoothness under complex paths in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0042] like Figure 1 As shown, the human specimen transposition vehicle following method based on camera vision in this embodiment includes the following steps:

[0043] S1. Collect environmental images in front of the human specimen transfer vehicle, convert the color images in the environmental images into grayscale images and calculate the gradient map, calculate the spot confidence based on the brightness characteristics of the pixels in the grayscale image and the gradient distribution characteristics in the gradient map, and generate a binarized mask based on the spot confidence to shield the specular reflection area.

[0044] Specifically, a binocular camera is deployed at the front of the human specimen transfer cart to acquire real-time color images of the target and its surrounding environment. Since the typical application scenarios are medical anatomy labs or specimen storage rooms, these environments are densely packed with stainless steel refrigerated cabinets and instrument tables. To effectively reduce the computational complexity of subsequent image processing and eliminate interference from multi-channel color information, this step converts the acquired color images into single-channel grayscale images.

[0045] After acquiring a single-channel grayscale image, the spatial gradient of the grayscale image is further calculated to accurately reflect the degree of spatial abrupt changes in pixel grayscale values. This embodiment employs the Sobel edge detection operator to calculate the spatial derivatives of the grayscale image in both the horizontal and vertical directions. Then, the sum of the squares of these two spatial derivatives is calculated, and the square root of the sum is taken to synthesize a complete gradient map. By constructing this gradient map, physical edge information where brightness changes drastically in the image can be accurately captured.

[0046] After completing basic image processing, the grayscale values ​​of each pixel in the grayscale image and the corresponding local gradient magnitudes in the gradient map are statistically analyzed to evaluate and calculate the confidence level of the light spot for each pixel. The unique specular reflection physical characteristics of stainless steel surfaces exhibit two extremely significant features in digital imaging: extremely high brightness values ​​and extremely steep edge gradients. Based on this physical-image mapping relationship, this step pre-sets grayscale and gradient thresholds. For any pixel in the image spatial pixel array, if its grayscale value in the grayscale image is greater than the preset grayscale threshold, and its local gradient magnitude in the gradient map is greater than the preset gradient threshold, then the pixel is determined to be a specular interference point. In this embodiment, the preset grayscale threshold is set within the high-brightness range of 240-255, and the preset gradient threshold is set to 500 based on the photosensitivity characteristics of the visual hardware. When a pixel simultaneously meets the condition that both of the above values ​​are greater than the corresponding preset threshold, it can be identified as a strong specular reflection point, that is, a specular interference point. Subsequently, the area containing the above specular interference point is identified as a high-spot confidence area.

[0047] Based on the pixel determination results above, a binary mask matrix with the same resolution as the original environment image is generated. In the logical settings of this matrix, the matrix value corresponding to the high-confidence area is forcibly assigned to 0, so as to completely block the visual signal of the specular reflection area, while the matrix value of the other normal areas without strong reflection is assigned to 1 and retained.

[0048] In the subsequent image feature point extraction steps, the underlying algorithm will be strictly restricted to operate within the unoccluded area where the binary mask matrix value is 1. This step, employing the aforementioned image acquisition and mask preprocessing techniques, can accurately and efficiently shield dense, bright spots excited by surgical shadowless lamps or strong indoor lighting on a smooth stainless steel surface during the initial processing stage of visual recognition. This data filtering method based on physical optical properties prevents the visual tracking algorithm from misjudging bright interference points in the environment as feature points of the target image, defining a highly pure image processing area for subsequent extraction of the human motion features of the actual target, and improving the target locking accuracy of the human specimen transfer vehicle in complex medical lighting environments.

[0049] S2. Extract image feature points in the region not obscured by the binarization mask, calculate the optical flow vector of the image feature points between consecutive frames, distinguish the real target from the false reflection based on the direction consistency of the optical flow vector and the perspective projection law, and remove abnormal feature points to lock the following target.

[0050] After generating the binarized mask and identifying the unoccluded areas, the AKAZE algorithm is used for feature detection in these areas. Compared to traditional image processing algorithms that rely on Gaussian blur, the AKAZE algorithm has the advantage of using nonlinear diffusion filtering to construct scale space for extracting image feature points. This nonlinear processing mechanism effectively suppresses environmental noise within the image while preserving the true edge shape and detailed structure of the object being measured to a great extent. This algorithm's characteristics are highly suitable for extracting complex texture features from the surfaces of metal equipment in scenarios such as medical anatomy labs, enabling more accurate and stable extraction of high-quality image feature points from the preserved effective areas, forming an initial feature point set.

[0051] After successfully extracting the set of image feature points, the spatial position changes of these feature points over time are tracked using the pyramid optical flow method. The system calculates the planar displacement of each image feature point between the current image frame and the previous adjacent image frame, thereby deriving the optical flow vector describing the motion state of the feature point. This optical flow vector is composed of horizontal and vertical motion components. In the actual operating environment of a human specimen transfer vehicle, the highly polished stainless steel refrigerated cabinet door surface will form a strong specular reflection effect. When the operator moves in front of the vehicle, not only are the features of the real operator reflected on the stainless steel surface, but also a false reflection of the moving person is simultaneously projected. In order to accurately isolate the real following target from the mixed set of feature points, this step introduces a comprehensive judgment based on the directional consistency of motion features and the perspective projection law in physical space.

[0052] According to kinematic analysis, the various image feature points on the surface of a real operator's body move as a whole in three-dimensional space. When mapped onto the two-dimensional image imaging plane, the optical flow vectors of these feature points should exhibit a high degree of directional consistency. For example, when the operator turns or translates laterally, the optical flow vectors of all feature points extracted from their torso will uniformly point to the corresponding side. Conversely, there is a significant depth difference between the imaging plane of the false reflection—the surface of the stainless steel equipment—and the actual physical plane of the real human body. Furthermore, the surface of the stainless steel equipment is often not an ideally flat surface, usually exhibiting slight curvature or metallic deformation. Based on perspective projection principles, this spatial depth difference and surface physical deformation will cause severe distortion, lateral slippage, or abnormal scaling of the optical flow vectors of the reflection projected onto it during imaging, thus causing its motion vectors to deviate from the overall motion consistency trend of the real target.

[0053] This step transforms the aforementioned physical laws into a mathematical screening and elimination model. First, it calculates the mean vector of the optical flow vectors of all extracted image feature points, and simultaneously calculates the standard deviation of this set of optical flow vector data. For any image feature point in the current feature point set, it calculates the Euclidean distance between the optical flow vector of each image feature point and the aforementioned mean vector. By comparing this Euclidean distance with a preset statistical threshold, the reasonableness of the motion state of each feature point is evaluated. The judgment logic here is based on the following mathematical relationship:

[0054]

[0055] In this relation, This represents the Euclidean distance between the optical flow vector and the mean vector of each image feature point, calculated from the given values. The mean vector representing the optical flow vectors of all image feature points. This represents the optical flow vector of a specific image feature point currently being calculated and determined. The standard deviation of the optical flow vector distribution of all image feature points is represented by the formula. This is a preset statistical threshold set according to the principles of statistical distribution. When the Euclidean distance of a certain image feature point calculated by substituting it into the formula is greater than the preset statistical threshold, it indicates that the motion trajectory of the image feature point deviates significantly from the overall motion trend of the subject in the image. Based on this calculation and comparison result, this step determines the image feature point whose Euclidean distance is greater than the preset statistical threshold as a false reflection or background noise, that is, an abnormal feature point, and removes it.

[0056] After the above screening, the data ultimately retained in the system's tracking list represents the true target feature group, free from various environmental interferences. Through this composite architecture combining nonlinear diffusion filtering algorithms and rigorous optical flow vector verification, the human specimen transposition vehicle system can accurately distinguish between real operators and false images on stainless steel equipment in medical equipment environments with strong reflections and complex curved surface mappings, ensuring the uniqueness and accuracy of the transposition vehicle's visual tracking target from the source of data input.

[0057] S3. Based on the number of valid feature points retained in the current frame and the average optical flow reprojection error of the valid feature points, the measured physical Euclidean distance is weighted to calculate the effective following distance after visual confidence weighting.

[0058] To prevent erroneous vehicle actions caused by sudden changes in visual signals, this step introduces a visual confidence assessment mechanism to calculate a weighted distance, which is a comprehensive correction for visual signal quality; this is the effective following distance. The formula for calculating the effective following distance is:

[0059]

[0060] In the formula, This represents the effective following distance after being weighted by visual confidence. This represents the physical Euclidean distance between the human specimen transfer vehicle and the target being followed, measured directly by the binocular camera. This indicates the number of valid feature points retained in the current frame after the preceding motion consistency check; This represents the feature quantity saturation constant, used to smooth out the impact of fluctuations in the number of feature points. It represents the average optical flow reprojection error of effective feature points, and its magnitude directly characterizes the instability of the visual tracking process and feature matching. This represents the error penalty coefficient, used to amplify the warning weight that errors bring to the system.

[0061] In vehicle safety following systems, a shorter distance between the vehicle and the target indicates a higher risk of collision. This formula directly reduces the quality evaluation metrics from the image processing level and transforms them into an intuitive distance control signal. When the number of effective feature points decreases sharply, meaning the system faces the risk of losing the target, the formula... The value of the term will decrease sharply; at the same time, when the running screen experiences severe shaking or feature matching errors cause the average optical flow reprojection error to increase, the value in the formula will decrease. The denominator of the term increases significantly, causing the value of the term to decrease substantially. Under these two conditions representing a decline in visual signal quality, the multiplication of the two decreased values ​​results in the calculated effective following distance being significantly smaller than the actual physical Euclidean distance currently measured by the binocular camera. Because the calculated effective following distance decreases drastically in value, the underlying control system receives a feedback signal that the distance has significantly decreased, determines that the target is very close to the vehicle, and immediately and actively triggers the system's deceleration or stopping protection logic.

[0062] By using this weighted calculation method that deeply couples visual confidence index with physical distance variable, the human specimen transposition vehicle can effectively shield visual distance jumps caused by external lighting or occlusion, avoiding the potential safety risk of the vehicle losing control and lurching forward due to visual signal attenuation or sudden loss.

[0063] S4. The real-time load mass is obtained through the pressure sensor in the hydraulic lifting circuit of the human specimen transfer vehicle, and the variable mass safe braking distance is calculated in combination with the current driving speed of the human specimen transfer vehicle; an exponential flexible speed command is generated based on the difference between the effective following distance and the variable mass safe braking distance, and the human specimen transfer vehicle is controlled to follow the target.

[0064] Because the overall mass of the human specimen transfer vehicle differs greatly between its unloaded state (containing only the vehicle's own weight) and its fully loaded state (containing specimens and preservatives), traditional fixed braking distance control strategies cannot meet the safety requirements of avoiding inertial forward lurch under heavy load conditions. Therefore, it is necessary to obtain the physical load in real time. This step involves reading the pressure sensor value installed in the hydraulic lifting circuit at the bottom of the hydraulic cylinder, and converting this fluid pressure value into the current real-time load mass according to Pascal's principle. Then, this real-time load mass is added to the vehicle's own weight to obtain the total mass of the human specimen transfer vehicle.

[0065] After obtaining the total vehicle mass, this step, combined with the current driving speed fed back in real time by the motor encoder, dynamically calculates the variable-mass safe braking distance that conforms to the current dynamic conditions. The formula for calculating the variable-mass safe braking distance is:

[0066]

[0067] The principle behind this formula is that since the kinetic energy of a moving vehicle is proportional to the square of its mass and speed, the braking system needs to do work to dissipate this energy. Therefore, the required physical safety braking distance for a vehicle is not a fixed value, but must be non-linearly positively correlated with the square of the vehicle's total mass and speed.

[0068] In the formula, Indicates the variable mass safety braking distance under the current operating conditions; This represents the total vehicle mass calculated based on real-time load mass. Indicates the current driving speed; This represents the kinetic braking coefficient, which reflects the actual energy dissipation capacity of the hydraulic braking system configured in the human specimen transfer vehicle; This represents the system response time constant, which includes the image processing delay of the vision system and the mechanical delay of the underlying hydraulic valve action; This represents a rigid safety margin, which ensures that no physical collision will occur under extreme circumstances where there are measurement deviations.

[0069] This dynamic model shows that when the total mass of the human specimen transshipment vehicle increases significantly due to the loading of a large number of specimens and liquids, the calculated variable-mass safe braking distance will increase accordingly. This ensures that, at the same travel speed, a fully loaded human specimen transshipment vehicle will trigger deceleration earlier than when unloaded, thus effectively compensating for the huge physical inertia brought about by heavy loads.

[0070] After determining the dynamic safety braking boundary, to eliminate the start-stop jerking sensation that is easily generated in hydraulic drive equipment under traditional linear proportional control, this step uses an exponential decay model to smooth speed changes and generate a target speed command. Based on the difference between the effective following distance (weighted by visual confidence) and the variable-mass safety braking distance, an exponential flexible speed command is generated, specifically satisfying the following formula:

[0071]

[0072] In the formula, This indicates the target speed command output to the motor controller in an exponential flexible speed command. This indicates the preset maximum cruising speed of the human specimen transfer vehicle; This represents the flexible damping length constant, which determines the smoothness of the system's deceleration curve; It is the base of the natural logarithm; This represents the effective following distance calculated through the preceding steps.

[0073] The underlying logic of this formula lies in implementing non-linear, smooth, piecewise control of the following distance. Specifically, when the effective following distance is greater than the variable-mass safe braking distance, the system is in a flexible deceleration range. Within this range, as the vehicle approaches the target, the difference between the effective following distance and the variable-mass safe braking distance gradually decreases. (The formula then continues with this explanation.) The terms gradually approach 1, thus making The calculated results and the final target speed command exhibit a gradual convergence and decrease trend. This exponential function's mathematical properties ensure that the vehicle maintains a sensitive response and high cruising speed when far from the target, while the rate of speed change becomes extremely gentle when approaching the target. This deceleration curve avoids system shocks caused by sudden closure of the hydraulic proportional valve, effectively preventing sloshing and violent splashing of the anti-corrosion liquid inside the container. When the effective following distance is less than or equal to the variable mass safe braking distance, it indicates that the vehicle has exceeded the safety boundary and entered the forced braking zone. To prevent a collision, the system directly bypasses the exponential decay calculation, forcibly setting the output target speed command to 0 and immediately triggering the vehicle's parking action. Based on this complete architecture that includes load and nonlinear speed regulation, the human specimen transposition vehicle can effectively handle various extreme mass changes and complex following trajectories.

[0074] The following combination Figures 2 to 4 The effects of implementing the present invention will be further explained.

[0075] like Figure 2 As shown in the figure, this diagram illustrates the process by which the underlying algorithm accurately distinguishes between the real target being followed and various visual disturbances. The diagram contains multiple sets of vector line segments with directional indicators and discretely distributed points. One set consists of vector line segments with highly consistent directions, representing the operator's real motion feature points that have been verified and ultimately locked by the algorithm. The other set consists of vector line segments with obvious lateral slippage and scaling distortion, representing false reflections projected onto a smooth stainless steel surface. These abnormal motion features are explicitly identified as disturbances by the system and removed. Furthermore, the scattered points in the diagram represent high-brightness noise points that were directly filtered out during the earlier generation of the binarization mask.

[0076] like Figure 3 As shown in the diagram, the figure contains two nested, elliptical safety boundary regions with significantly different coverage sizes. The smaller inner region represents the safe braking range when the human specimen transfer vehicle is unloaded, containing only its own weight; the significantly expanded outer region represents the safe braking range when the human specimen transfer vehicle is fully loaded with a large number of specimens and preservatives. This dynamic boundary change intuitively reflects the actual control effect of the variable-mass safety braking model based on the kinetic energy theorem. That is, as the real-time load mass increases, the system-calculated safe braking distance automatically extends, and the safety warning range expands synchronously to effectively compensate for the surge in physical inertia.

[0077] like Figure 4As shown in the figure, the dotted line trajectory in the center represents the operator's baseline walking trajectory. Traditional visual tracking technology, when affected by high reflectivity in the surrounding environment and lacking inertial compensation, generates vehicle movement trajectories that exhibit dense oscillations and irregular sawtooth fluctuations, easily causing vehicle shaking. In contrast, the vehicle movement trajectory generated by this invention exhibits extremely high smoothness and continuity, stably conforming to the operator's baseline walking route throughout. This demonstrates the excellent effectiveness of the technical solution of this invention in eliminating system dynamic mismatch, avoiding hydraulic drive start-stop jerks, and ensuring the smooth and safe transportation of chemical anti-corrosion liquids.

[0078] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A camera vision-based method for following a human specimen transposition vehicle, characterized in that, include: The environmental image in front of the human specimen transfer vehicle is acquired, the color image in the environmental image is converted into a grayscale image and the gradient map is calculated. The confidence of the light spot is calculated based on the brightness characteristics of the pixels in the grayscale image and the gradient distribution characteristics in the gradient map. A binary mask is generated based on the confidence of the light spot to shield the specular reflection area. Image feature points are extracted in areas not obscured by binarization masks. The optical flow vectors of the image feature points are calculated between consecutive frames. Based on the consistency of the direction of the optical flow vectors and the perspective projection rules, real targets and false reflections are distinguished. Abnormal feature points are removed to lock onto the following target. Based on the number of valid feature points retained in the current frame and the average optical flow reprojection error of the valid feature points, the measured physical Euclidean distance is weighted to calculate the effective following distance after visual confidence weighting. The calculation formula is as follows: In the formula, This represents the effective following distance after weighting by visual confidence. Represents the physical Euclidean distance; This indicates the number of valid feature points retained in the current frame; Indicates the saturation constant of the characteristic quantity; The average optical flow reprojection error represents the effective feature points. Indicates the error penalty coefficient; The real-time load mass is obtained by a pressure sensor in the hydraulic lifting circuit of the human specimen transfer vehicle. Combined with the current travel speed of the human specimen transfer vehicle, the variable-mass safe braking distance is calculated using the following formula: In the formula, Indicates the variable mass safety braking distance; This represents the total vehicle mass calculated based on real-time load mass. Indicates the current speed of the human specimen transfer vehicle; Indicates the kinetic energy braking coefficient; Indicates the system response time constant; Indicates a rigid safety margin; An exponential flexible speed command is generated based on the difference between the effective following distance and the variable mass safe braking distance to control the human specimen transfer vehicle to follow the target's movement, including: In the formula, This indicates the target speed command output to the motor controller in an exponential flexible speed command. This indicates the preset maximum cruising speed; This represents the flexible damping length constant; It is the base of the natural logarithm; When the effective following distance is less than or equal to the variable mass safe braking distance, the target speed command value is 0.

2. The human specimen transposition vehicle following method based on camera vision according to claim 1, characterized in that, The step of calculating the confidence level of the light spot based on the brightness characteristics of pixels in the grayscale image and the gradient distribution characteristics in the gradient map includes: Statistically analyze the grayscale value and local gradient magnitude of each pixel in the grayscale image; When the gray value of the pixel is greater than a preset gray value threshold and the local gradient magnitude is greater than a preset gradient threshold, the pixel is determined to be a speckle interference point, and the area containing the speckle interference point is determined to be a speckle confidence area.

3. The human specimen transposition vehicle following method based on camera vision according to claim 1, characterized in that, The extraction of image feature points uses the AKAZE algorithm, which utilizes nonlinear diffusion filtering to construct a scale space.

4. The human specimen transposition vehicle following method based on camera vision according to claim 3, characterized in that, The process of removing abnormal feature points includes: calculating the mean vector of the optical flow vectors of all extracted image feature points; calculating the Euclidean distance between the optical flow vector of each image feature point and the mean vector; and identifying image feature points whose Euclidean distance is greater than a preset statistical threshold as abnormal feature points and removing them.

5. The human specimen transposition vehicle following method based on camera vision according to claim 1, characterized in that, The method of obtaining real-time load mass through the pressure sensor in the hydraulic lifting circuit of the human specimen transfer vehicle includes: reading the fluid pressure value output by the pressure sensor installed at the bottom of the hydraulic cylinder on the human specimen transfer vehicle, and converting the fluid pressure value into the real-time load mass according to Pascal's principle.

6. The human specimen transposition vehicle following method based on camera vision according to claim 1, characterized in that, The total mass of the vehicle includes the vehicle's own weight and the mass of the specimen and liquid calculated from the readings of the pressure sensor. The kinetic braking coefficient is a constant set based on the energy dissipation capacity of the hydraulic braking system configured on the human specimen transfer vehicle.

7. The method for following a human specimen transposition vehicle based on camera vision according to claim 1, characterized in that, The gradient map calculation includes: using the Sobel edge detection operator to calculate the spatial derivatives of the grayscale image in the horizontal and vertical directions respectively; obtaining the sum of squares of the spatial derivatives in the horizontal and vertical directions, and performing a square root operation on the sum of squares to synthesize the gradient map.

Images