A method for reducing forklift collision risk based on LSTM

Abstract

The application provides a method for reducing the collision risk of a forklift based on LSTM, and relates to the technical field of industrial vehicle safety. The method for reducing the collision risk of a forklift comprises the following steps: S1, monitoring the state and position of pedestrians around the vehicle, predicting the trajectory of the pedestrians, and setting a safety line and a braking line; S2, judging whether the pedestrians are located on the driving path of the vehicle; S3, if the pedestrians are located on the driving path, controlling the vehicle to perform corresponding actions according to the motion direction relationship and distance relationship between the pedestrians and the vehicle; and S4, if the pedestrians are not located on the driving path, analyzing the position relationship between the pedestrians and the safety line and the braking line under different behaviors of the pedestrians in combination with the predicted trajectory of the pedestrians, and controlling the vehicle to perform corresponding actions. The application sets double dynamic boundaries, judges whether the pedestrians are located on the driving path, performs differentiated multi-layer grading decisions, significantly reduces the false positive rate while ensuring safety, and improves the accuracy and foresight of early warning, and the intelligence and efficiency of system decision.

Description

Technical Field

[0001] This invention relates to the field of industrial vehicle safety technology, and in particular to a method for reducing the risk of forklift collisions based on LSTM. Background Technology

[0002] With the rapid development of industrial logistics and warehousing management, forklifts, as core handling equipment, are widely used in ports, workshops, warehouses and other scenarios. However, in complex environments with mixed traffic of people and vehicles, collisions between forklifts and pedestrians occur frequently, becoming a major hidden danger to industrial safety. According to accident statistics, forklift accidents account for 97% of all accidents involving special-purpose motor vehicles on site, with collisions and crushing accidents accounting for more than 60%. Blind spots, speeding and human error are the main causes.

[0003] Traditional forklift safety protection technology has significant defects: it relies on passive protection methods such as physical crash bars and static audible and visual alarms, and lacks the ability to predict and actively intervene in pedestrian behavior. Existing crash protection systems based on fixed area monitoring (such as the front square alarm zone) can provide basic warnings of positive risks, but cannot effectively cover the blind spots on the sides of the forklift, resulting in a serious lag in warnings of pedestrians approaching from the side.

[0004] Chinese Patent Application No. 2025109293311 discloses a forklift and pedestrian trajectory prediction based on multi-camera fusion, relating to the field of forklift and pedestrian trajectory prediction and collision warning technology. This invention acquires panoramic and regional images within a factory building, establishes panoramic and regional detection models, and detects the movement status and pedestrian behavior characteristics of forklifts and pedestrians, as well as driver driving data and forklift characteristics. By establishing a pedestrian trajectory prediction model, it predicts pedestrian walking paths based on basic pedestrian prediction data, movement status, and current work content. Similarly, by establishing a forklift trajectory prediction model, it plans forklift handling routes based on basic forklift prediction data, movement status, and current loading / unloading content. It assesses the driver's compatibility with the forklift by providing real-time driving safety scores and forklift status safety scores. Finally, it sets forklift warning times and provides collision warnings and methods based on the predicted trajectories of pedestrians and forklifts.

[0005] Similar to the existing technologies mentioned above, although some companies have tried to improve the accuracy of early warning by introducing skills such as UWB positioning and AI visual recognition, for example, by using UWB tags to achieve centimeter-level distance measurement or by using AI cameras to identify pedestrians, existing solutions still have many bottlenecks: the system can only respond to the established fact that a pedestrian has "entered" the monitoring area, and it is difficult to solve the problem of insufficient reaction time when a pedestrian suddenly cuts into the path from the side; most systems lack a graded intervention mechanism and cannot dynamically adjust the response strategy according to the pedestrian's position, direction of movement and risk level, resulting in frequent false braking or insufficient intervention; and it is difficult to achieve differentiated protection for special groups such as the elderly and children.

[0006] Given the aforementioned challenges and the increasingly urgent need for forklift operation safety, how to overcome the key technical bottlenecks of traditional collision avoidance systems, such as delayed warnings, high false alarm rates, and weak scenario adaptability, through the integration of intelligent perception, trajectory prediction, and dynamic decision-making, has become a core issue that the industry urgently needs to address.

[0007] Therefore, it is necessary to invent a method based on LSTM to reduce the risk of forklift collisions and solve the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a method for reducing the risk of forklift collisions based on LSTM, which aims to solve the problem of insufficient reaction time when pedestrians suddenly cut into the path from the side.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a method for reducing forklift collision risk based on LSTM, comprising: S1, real-time monitoring of the status and position of pedestrians around the vehicle, and predicting the pedestrian's movement trajectory based on an LSTM model; and setting a safety line and a braking line around the vehicle, with the braking line closer to the vehicle than the safety line; S2, determining whether the pedestrian is located on the vehicle's current driving path; S3, if the pedestrian is located on the driving path, controlling the vehicle to perform corresponding actions based on the relationship between the pedestrian's and the vehicle's movement direction and distance; S4, if the pedestrian is not on the driving path, combining the pedestrian's predicted movement trajectory, analyzing the positional relationship between the pedestrian and the safety line and braking line under different behavioral assumptions, and controlling the vehicle to perform corresponding actions.

[0010] Preferably, step S3 includes: if the pedestrian and vehicle are moving in the same direction and are outside the safety line, control the vehicle to drive normally; if the pedestrian and vehicle are moving in opposite directions, or are moving in the same direction but are inside the safety line, control the vehicle to slow down and issue a warning signal; if the pedestrian is inside the braking line, control the vehicle to brake.

[0011] Preferably, step S4 includes: if the pedestrian is outside the safety line and is predicted to approach the vehicle, controlling the vehicle to slow down and issuing a warning; if the pedestrian is outside the safety line and is predicted not to approach the vehicle, controlling the vehicle to drive normally; if the pedestrian is inside the safety line but is predicted not to approach the vehicle, and is outside the brake line, controlling the vehicle to slow down and issuing a warning, and is inside the brake line, controlling the vehicle to brake; if the pedestrian is inside the safety line and is predicted to approach the vehicle, controlling the vehicle to brake.

[0012] Preferably, when a pedestrian is outside the safety line and is predicted to approach the vehicle, the following judgment is further performed: if the pedestrian's movement direction is opposite to the vehicle's driving direction, and it is predicted that the pedestrian will not enter the braking line after accelerating, the vehicle is controlled to drive normally; otherwise, the vehicle is controlled to slow down and issue a warning.

[0013] Preferably, monitoring pedestrian status in step S1 includes: identifying pedestrians' body shape, movements, and facial features through a visual sensor; distinguishing pedestrian types based on the identification results; and initiating differentiated protection strategies for special pedestrian types.

[0014] Preferably, the differentiated protection includes: automatically expanding the preset distance between the safety line and the braking line when the person is identified as a child; switching the warning signal to a soft light effect and a gentle voice prompt when the person is identified as an elderly person; and restoring the default settings when the person is not a special group.

[0015] Preferably, the process of identifying pedestrians through a visual sensor includes: acquiring pedestrian images, extracting the body shape features of the pedestrian images using a convolutional neural network, analyzing the pedestrian's gait using skeletal key point tracking technology, and comprehensively judging the pedestrian's age stage by combining facial features.

[0016] Preferably, the step of predicting the pedestrian's movement trajectory in step S1 includes: obtaining the relative position sequence of the pedestrian relative to the vehicle; fusing the vehicle's own motion state to calculate the absolute coordinate sequence of the pedestrian; inputting the absolute coordinates of multiple time series frames into the LSTM network, and outputting the probability distribution of the pedestrian's next movement trajectory.

[0017] Preferably, in step S1, the distance threshold between the braking line and the safety line is dynamically calculated based on the vehicle's maximum braking distance, the pedestrian's average moving speed, and the system response delay.

[0018] Preferably, in step S1, the real-time position, velocity and acceleration data of pedestrians are fused by an LSTM model to generate dynamic behavior pattern classification and trajectory prediction. The pedestrian behavior patterns include walking, running and standing still, which are achieved by classifying pedestrian gait frequency and displacement change rate through an LSTM model.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves accurate prediction of pedestrian trajectories using an LSTM model, enabling proactive assessment of lateral risks and overcoming the lag issue of traditional forward square warning zones. By setting dual dynamic boundaries of safety lines and braking lines, and combining this with the determination of whether a pedestrian is within the driving path, differentiated multi-level hierarchical decision-making is implemented, significantly reducing the false alarm rate while ensuring safety. Through a zoned monitoring strategy for the forward and lateral warning areas, and adaptive safety strategies for special pedestrian types such as children and the elderly, comprehensive and humane protection is achieved for the vehicle's surrounding environment and different groups of people. Ultimately, through the synergistic effect of prediction, zoning, hierarchical, and adaptive technologies, the accuracy and foresight of warnings, the intelligence and efficiency of system decision-making, and the overall safety level in complex scenarios with mixed pedestrian and vehicle traffic are significantly improved. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0021] Figure 2 This is a schematic diagram of a sub-process when a pedestrian is located on the vehicle's travel path in this invention.

[0022] Figure 3 This is a schematic diagram of the sub-process when the pedestrian is not on the vehicle's travel path in this invention.

[0023] Figure 4 This is a schematic diagram of a sub-process for initiating differentiated protection for special pedestrians in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0025] To improve the vehicle's reaction time when pedestrians enter its path from the side, such as Figures 1 to 3 As shown, this invention proposes a method for reducing forklift collision risk based on LSTM, including: S1 monitors the status and position of pedestrians around the vehicle in real time and predicts the pedestrian's movement trajectory based on the LSTM model; and sets safety lines and brake lines around the vehicle, with the brake lines closer to the vehicle than the safety lines. S2, determine whether the pedestrian is on the vehicle's current driving path; S3, if the pedestrian is on the driving path, the vehicle will be controlled to perform corresponding actions based on the relationship between the pedestrian's and the vehicle's direction of movement and the distance between the pedestrian and the vehicle; S4. If the pedestrian is not on the driving path, the vehicle is controlled to perform corresponding actions based on the predicted movement trajectory of the pedestrian and the positional relationship between the pedestrian and the safety line and the braking line under different behavioral assumptions.

[0026] The step of predicting the pedestrian's movement trajectory in step S1 includes: S11, obtain the relative position sequence of pedestrians relative to vehicles; S12, integrates the vehicle's own motion state to calculate the absolute coordinate sequence of pedestrians; S13: Input the absolute coordinates of multiple time series frames into the LSTM network and output the probability distribution of the pedestrian's next movement trajectory.

[0027] It should be noted that in step S11, the relative position and relative distance between the pedestrian and the vehicle can be determined by the visual sensor and the distance sensor. Based on the pixel points and distance information of the pedestrian relative to the vehicle, multiple coordinates (x1, y1), (x2, y2), (x3, y3), etc., of the human body relative to the forklift can be generated.

[0028] In step S12, the vehicle's speed and direction are determined by the vehicle system. The vehicle's speed and direction can be determined based on the gyroscope / CAN data set by the vehicle itself, or by combining the on-board inertial measurement unit and wheel speed encoder. The vehicle's motion state is estimated in real time and accurately through the Kalman filter algorithm. The vehicle's position is combined with the pedestrian's position, and an absolute coordinate system is generated based on the ground. The absolute coordinates (x1^y1), (x2^y2), and (x3^y3) are generated by combining the pedestrian's relative coordinates.

[0029] In step S13, the corrected coordinates are calculated using an LSTM model to output the pedestrian trajectory sequence {T1(x1^y1), T2(x2^y2), T3(x3^y3), Tn(xn^yn)} for future times, where Tn(xn^yn) is the predicted value of the pedestrian trajectory.

[0030] In this embodiment, in step S1, the real-time position, velocity and acceleration data of pedestrians can be fused by an LSTM model to generate dynamic behavior pattern classification and trajectory prediction. The pedestrian behavior patterns include walking, running and standing still. The classification of pedestrian gait frequency and displacement change rate is achieved by the LSTM model.

[0031] An attention mechanism can be introduced into the LSTM network to dynamically focus on the key time step information in the pedestrian's historical trajectory that is most relevant to future movement during prediction. For example, when a pedestrian shows signs of suddenly turning, the attention mechanism will give higher weight to the trajectory features at that moment, thereby predicting the turning intention earlier and more accurately, and further improving the foresight of the warning.

[0032] The visual sensor can use a high dynamic range global shutter camera to solve the problem of image blurring or overexposure caused by drastic changes in lighting in the warehouse (such as entering and exiting the garage, and the shadow of the shelves), ensuring that pedestrian images can be captured stably under different lighting conditions; the distance sensor can integrate solid-state LiDAR and millimeter-wave radar. LiDAR provides high-precision point cloud data for contour recognition and distance measurement, while millimeter-wave radar has a strong anti-environment interference capability and can still work stably under harsh conditions where the visual sensor fails, making up for the limitations of a single sensor.

[0033] In step S1, the vehicle's monitorable area is divided into a square area in front and a side warning area based on the vehicle's direction of travel. When a pedestrian appears in the square area in front, step S3 determines the pedestrian's risk level based on the relationship between the pedestrian's and the vehicle's movement direction and distance, and issues a warning and determines the vehicle's action. When a pedestrian appears in the side warning area, step S4 analyzes the pedestrian's positional relationship with the safety line and brake line under different behavioral assumptions based on the pedestrian's predicted movement trajectory, determines the pedestrian's risk level, issues a warning, and determines the vehicle's action.

[0034] In step S1, the brake line and safety line are set according to the risk posed by pedestrians. The distance threshold between the brake line and the safety line is dynamically calculated based on the vehicle's maximum braking distance, the average moving speed of the pedestrian, and the system response delay. The brake line is set according to the pedestrian's limit acceleration, so that even if the pedestrian suddenly accelerates, the vehicle can complete the braking before contacting the pedestrian. Moreover, the safety line and brake line in the vehicle's forward direction are farther away from the vehicle than the side of the vehicle. The underlying logic of controlling the vehicle based on the brake line settings controls the vehicle to brake as soon as the pedestrian enters the brake line.

[0035] The distance threshold between the safety line and the braking line can be controlled using fuzzy logic. This method not only considers physical parameters but also takes environmental complexity coefficients (such as shelf density and aisle width) and driver behavior scores (such as historical operational stability) as input variables. It can set a slightly larger safety margin for cautious drivers in complex environments, while setting a relatively compact but still safe threshold for aggressive drivers or in open environments, thus achieving a precise balance between safety and efficiency.

[0036] In this embodiment, after the judgment in step S2, if the pedestrian is located on the vehicle's travel path, the process proceeds to step S3, which is the same-direction movement and safety line judgment sub-process. The specific steps of step S3 include: S31, If ​​a pedestrian and a vehicle are moving in the same direction and are outside the safety line, control the vehicle to drive normally; S32, If pedestrians and vehicles move in opposite directions, or if they move in the same direction but are within the safety line, control the vehicle to slow down and issue a warning signal; S33: If the pedestrian is inside the brake line, control the vehicle to brake.

[0037] When a pedestrian is on the driving path, the system determines whether the pedestrian's direction of travel is the same as that of the vehicle based on the real-time monitored pedestrian position and the pedestrian trajectory predicted by the LSTM model. When the pedestrian and vehicle are moving in the same direction, the system monitors the distance between the pedestrian and vehicle in real time. When the distance between the pedestrian and vehicle decreases to a certain distance, that is, when the pedestrian enters the vehicle's safety line, the system controls the vehicle to slow down and issues a warning. If the pedestrian moves away from the vehicle to outside the safety line, the system controls the vehicle to resume normal driving. If the pedestrian does not move away after the vehicle issues a warning, the system controls the vehicle to brake when the pedestrian enters the braking line.

[0038] When it is determined that the pedestrian's direction of travel is opposite to that of the vehicle, the vehicle is controlled to slow down directly and issue a warning to prevent a collision due to the high relative speed when the pedestrian and the vehicle are traveling towards each other; when the pedestrian is on the vehicle's path and is moving outside the vehicle's path, if the pedestrian is within the vehicle's safety line, the vehicle is controlled to slow down and issue a warning; if the pedestrian is not within the vehicle's safety line, the vehicle travels normally, and after the pedestrian enters the side area of ​​the vehicle, step S4 is entered to analyze the collision risk of the pedestrian and control the vehicle's actions.

[0039] In this embodiment, after step S2, if the pedestrian is not on the vehicle's travel path, the process proceeds to step S4, which is the sub-process for classifying safety lines and brake lines. The specific steps of step S4 include: If a pedestrian is outside the safety line and is predicted to approach the vehicle, slow down the vehicle and issue a warning. If the pedestrian is outside the safety line and is not expected to approach the vehicle, control the vehicle to drive normally. If a pedestrian is inside the safety line but is not expected to approach the vehicle, and is outside the brake line, control the vehicle to slow down and issue a warning; if the pedestrian is inside the brake line, control the vehicle to brake. If a pedestrian is inside the safety line and is predicted to approach the vehicle, control the vehicle to brake.

[0040] It should be noted that after step S2 determines the position of the pedestrian relative to the vehicle safety line, the pedestrian trajectory predicted by the LSTM model is used to determine whether the pedestrian will approach the vehicle. When the pedestrian is inside the safety line, the predicted trajectory of the pedestrian will approach the vehicle, and then the relative position of the pedestrian and the braking line is determined.

[0041] In this embodiment, when a pedestrian is outside the safety line and is predicted to approach a vehicle, the following judgment is further performed: If the pedestrian moves in the opposite direction to the vehicle's direction of travel, and it is predicted that the pedestrian will not enter the braking line after accelerating, control the vehicle to drive normally. Otherwise, control the vehicle to slow down and issue a warning.

[0042] It should be noted that when a pedestrian approaches a vehicle outside the safety line and moves towards the rear of the vehicle, the LSTM model predicts the pedestrian's maximum acceleration and direction change path. If it is determined that the pedestrian will not enter the vehicle's braking line, the vehicle continues to travel at the original speed; if it is determined that the pedestrian will enter the vehicle's braking line, the vehicle is controlled to decelerate and a warning is issued.

[0043] When a pedestrian is outside the safety line, the system updates their motion vector and predicts their trajectory in real time. If it determines that there is a risk of collision with the vehicle, it triggers a deceleration mechanism in advance before the pedestrian crosses the safety line. If, after crossing the safety line, the system determines that there is a risk of the pedestrian continuing to approach the vehicle based on the predicted trajectory, it controls the vehicle to perform emergency braking in advance before the pedestrian touches the brake line, ensuring a complete stop before contact. The entire process is monitored in real time by the sensor fusion module, which combines high-precision maps and dynamic environment modeling to ensure accurate and reliable decision-making.

[0044] In this embodiment, as Figure 3 As shown, the detailed steps of step S4 can be summarized as follows: S41, determine whether the pedestrian is outside the preset safety line. If the result is yes, proceed to S42. If the result is no, proceed to S45. S42, Based on the predicted movement trajectory of the pedestrian, determine whether the pedestrian will approach the vehicle. If the result is yes, proceed to S43; if the result is no, proceed to S48. S43, Based on the predicted movement trajectory of the pedestrian, determine whether the pedestrian is moving in the opposite direction to the direction of vehicle travel. If the determination result is yes, proceed to S44; if the determination result is no, proceed to S47. S44, Based on the predicted movement trajectory of the pedestrian, determine whether the pedestrian will enter the preset braking line if the movement accelerates. If the determination result is yes, proceed to S47; if the determination result is no, proceed to S48. S45, Based on the predicted movement trajectory of the pedestrian, determine whether the pedestrian will approach the vehicle. If the result is yes, proceed to S49; if the result is no, proceed to S46. S46, determine whether the pedestrian is outside the preset brake line. If the result is yes, proceed to S47; if the result is no, proceed to S49. S47, control the vehicle to slow down and issue a warning, then return to S1; S48, control the vehicle to drive normally, return to S1; S49, control the vehicle braking, return to S1.

[0045] In this embodiment, by setting dual safety boundaries of safety lines and braking lines, and based on whether a pedestrian is on the path and the predicted trajectory, a differentiated, multi-level decision-making process is executed. For pedestrians on the path, the judgment is made in combination with their direction of movement (same direction / opposite direction), avoiding the limitations of a single distance threshold. This ensures safety while reducing unnecessary emergency braking and improving operational efficiency. For pedestrians outside the path (to the side), not only is the current position considered, but also the extreme behavior of the pedestrian (such as sudden acceleration) is hypothesized and analyzed. When it is predicted that the pedestrian will not cross the braking line even if accelerating, the vehicle does not decelerate or issue a warning, which improves work efficiency, greatly reduces the false alarm rate, and ensures the accuracy and necessity of the warning. Example 2

[0046] In actual use, there are instances where behavioral differences among special groups such as the elderly and children are not taken into account (e.g., children suddenly running, the elderly being sensitive to bright light and sound), making it difficult to achieve differentiated protection.

[0047] To solve the above-mentioned technical problems, in another embodiment of the present invention, such as Figures 1 to 4 As shown, monitoring pedestrian status in step S1 includes the following steps: Visual sensors are used to identify pedestrians' body shape, movements, and facial features. Based on the identification results, pedestrian types are distinguished, and differentiated protection strategies are initiated for special pedestrian types.

[0048] Differential protection includes: automatically extending the preset distance between the safety line and the brake line when a child is identified; switching the warning signal to a soft light effect and a gentle voice prompt when an elderly person is identified; and restoring the default settings when not a special group of people.

[0049] The process of identifying pedestrians using visual sensors includes: acquiring pedestrian images, extracting the body shape features of pedestrian images using convolutional neural networks, analyzing pedestrian gait movements using skeletal keypoint tracking technology, and comprehensively judging the age stage of pedestrians by combining facial features.

[0050] In this embodiment, as Figure 4 As shown, step S1 incorporates the following steps to change the risk scope and alert method based on pedestrian category: S51, to monitor pedestrians’ body shape, movements and appearance; S52, determine whether the person is a child based on their body shape. If the result is yes, proceed to S53; if the result is no, proceed to S54. S53, the control safety line and brake line are in an outward expansion state, enter S57; S54. Determine whether the person is an elderly person based on their actions or appearance. If the determination is yes, proceed to S55. If the determination is no, proceed to S56. S55, when slowing down the vehicle, change the warning methods such as horn and flashing lights that could be stimulating to the elderly to gentle, clear voice and soft, slowly flashing lights that can avoid startling them, and move to S57. S56, restore the changes that occurred in steps S53 and S55 to the normal working state, and proceed to S57; S57, return to the main process and continue with step S1.

[0051] In particular, since children's movement trajectories are difficult to predict and children are often active and move quickly, step S53 expands the safety line and braking line outward to prevent the vehicle from reacting in advance when children move quickly outside the predicted trajectory, thus reducing the risk of collision. Since the elderly are easily startled and may have accidents when subjected to sudden or strong stimuli, when monitoring the presence of the elderly, step S55 changes the warning methods that are likely to stimulate the elderly, such as horns and flashing lights, to calm voices and soft light effects to avoid startling the elderly when they approach the vehicle.

[0052] It should be noted that this embodiment takes into account the differentiated protection strategies for special groups such as the elderly and children, reflecting the humanized and refined design of the system. For children, the system uses a convolutional neural network to identify their body shape characteristics and automatically extends the safety line and braking line outward. This is because children's behavior is more sudden and their shorter stature makes them more likely to be in blind spots, requiring earlier warnings and longer braking distances.

[0053] For the elderly, the system identifies them by analyzing facial features and gait, and adjusts the warning method from rapid sirens and bright flashing lights that may cause fright to a gentle voice prompt and soft, slowly flashing lights; while ensuring the warning effect, it avoids secondary accidents caused by sudden stimulation that may cause the elderly to panic.

[0054] In this embodiment, by implementing the method of "LSTM-based trajectory prediction combined with hierarchical decision-making based on the dual dynamic boundaries of safety lines and braking lines," and coordinating with the zoned monitoring strategy of the square area in front of the vehicle and the side warning area, the core problems of traditional forklift collision avoidance systems—such as delayed side pedestrian warnings, high false alarm rates, and insufficient protection for special groups—are systematically solved. This significantly improves the foresight and decision-making accuracy of collision risk warnings. Based on predicted trajectories and extreme behavior simulations (such as sudden acceleration), corresponding protective mechanisms can be triggered in advance before pedestrians enter the danger zone, essentially eliminating side blind spot risks, greatly reducing the system's false alarm rate, and significantly improving operational continuity. Through the synergistic effect of the LSTM prediction module, the zoned monitoring module, and the differentiated decision-making logic, not only is comprehensive and intelligent management of 360-degree collision risks around the vehicle achieved, but also the safety zone and warning methods can be adaptively adjusted for special pedestrians such as children and the elderly. Ultimately, in complex mixed pedestrian and vehicle environments, significant improvements in safety, operational efficiency, and humanistic care are achieved simultaneously.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for reducing forklift collision risk based on LSTM, characterized in that, include: S1 monitors the status and location of pedestrians around the vehicle in real time and predicts the pedestrian's movement trajectory based on the LSTM model; Safety lines and brake lines are set around the vehicle, with the brake lines closer to the vehicle than the safety lines. S2, determine whether the pedestrian is located on the current driving path of the vehicle; S3, if the pedestrian is located on the driving path, the vehicle is controlled to perform corresponding actions according to the relationship between the pedestrian's and the vehicle's direction of movement and distance. S4. If the pedestrian is not on the driving path, the positional relationship between the pedestrian and the safety line and the braking line under different behavioral assumptions is analyzed based on the pedestrian's predicted movement trajectory, and the vehicle is controlled to perform corresponding actions.

2. The method for reducing forklift collision risk according to claim 1, characterized in that, Step S3 includes: If a pedestrian and a vehicle are moving in the same direction and are outside the safety line, control the vehicle to drive normally. If pedestrians and vehicles are moving in opposite directions, or if they are moving in the same direction but are within the safety line, the vehicle will be slowed down and a warning signal will be issued. If the pedestrian is inside the brake line, control the vehicle to brake.

3. The method for reducing forklift collision risk according to claim 1, characterized in that, Step S4 includes: If a pedestrian is outside the safety line and is predicted to approach the vehicle, slow down the vehicle and issue a warning. If the pedestrian is outside the safety line and is not expected to approach the vehicle, control the vehicle to drive normally. If a pedestrian is inside the safety line but is not expected to approach the vehicle, and is outside the brake line, control the vehicle to slow down and issue a warning; if the pedestrian is inside the brake line, control the vehicle to brake. If a pedestrian is inside the safety line and is predicted to approach the vehicle, control the vehicle to brake.

4. The method for reducing forklift collision risk according to claim 3, characterized in that, When a pedestrian is outside the safety line and is predicted to approach a vehicle, the following judgment is further performed: If the pedestrian moves in the opposite direction to the vehicle's direction of travel, and it is predicted that the pedestrian will not enter the braking line after accelerating, control the vehicle to drive normally. Otherwise, control the vehicle to slow down and issue a warning.

5. The method for reducing forklift collision risk according to claim 1, characterized in that, Monitoring pedestrian status in step S1 includes: Visual sensors are used to identify pedestrians' body shape, movements, and facial features. Based on the identification results, pedestrian types are distinguished, and differentiated protection strategies are initiated for special pedestrian types.

6. The method for reducing forklift collision risk according to claim 5, characterized in that, The differentiated protection includes: When a child is identified, the safety line and the brake line are automatically extended by a preset distance. When the system identifies an elderly person, it switches the warning signal to a soft light effect and a gentle voice prompt. Restore default settings for non-special populations.

7. The method for reducing forklift collision risk according to claim 6, characterized in that, The process of identifying pedestrians using visual sensors includes: acquiring pedestrian images, extracting the body shape features of the pedestrian images using a convolutional neural network, analyzing the pedestrian's gait using skeletal key point tracking technology, and comprehensively judging the pedestrian's age stage by combining facial features.

8. The method for reducing forklift collision risk according to claim 1, characterized in that, The steps in step S1 for predicting the pedestrian's movement trajectory include: Obtain the relative position sequence of pedestrians relative to vehicles; By integrating the vehicle's own motion state, the absolute coordinate sequence of pedestrians is calculated; The absolute coordinates of multiple time series frames are input into the LSTM network, which outputs the probability distribution of the pedestrian's next movement trajectory.

9. The method for reducing forklift collision risk according to claim 1, characterized in that, In step S1, the distance threshold between the braking line and the safety line is dynamically calculated based on the vehicle's maximum braking distance, the pedestrian's average moving speed, and the system response delay.

10. The method for reducing forklift collision risk according to claim 1, characterized in that, In step S1, the real-time position, velocity and acceleration data of pedestrians are fused by an LSTM model to generate dynamic behavior pattern classification and trajectory prediction. The pedestrian behavior patterns include walking, running and standing still. The classification of pedestrian gait frequency and displacement change rate is achieved by an LSTM model.

Images