1333 results about "Collision detection" patented technology

The invention provides a vehicle surrounding obstacle detection method and system based on a multi-modal sensor, and relates to the technical field of intelligent driving, and the method comprises the steps: obtaining ultrasonic radar point clouds and millimeter wave radar point clouds at two sides of a vehicle, and image data at the rear side of the vehicle; the radar point clouds are clustered, and the spatial position, the movement speed and the reflection section characteristics of an obstacle are extracted; performing target detection on the image data, and outputting to obtain a bounding box and texture features of an obstacle; generating preliminary multi-modal data, inputting the multi-modal data into a collision detection model, respectively entering a radar branch network and a visual branch network to extract point cloud spatial-temporal features and image semantic features, and performing weighted fusion on the point cloud spatial-temporal features and the image semantic features through a Transform mechanism to obtain a collision detection result; outputting the three-dimensional position, the movement track, the type label and the collision risk probability of the obstacle; and when the collision risk probability exceeds a preset threshold value, triggering a corresponding early warning or control instruction.

The invention relates to the technical field of mechanical arm obstacle avoidance path planning, in particular to a multi-degree-of-freedom mechanical arm obstacle avoidance path planning method based on three-dimensional reconstruction, which comprises the following steps: step 1, three-dimensional environment perception and dynamic modeling; preferentially offsetting and expanding the near-obstacle nodes towards the concave area or the hole center to generate a candidate node set; and 4, three-dimensional grid collision verification and safe path correction are conducted, specifically, the working space of the mechanical arm is divided into three-dimensional voxel grids, and collision detection is achieved by judging whether path nodes fall into obstacle object elements or not. According to the method, the laser radar and the depth camera are adopted to synchronously collect data through hardware triggering, statistical filtering denoising and three-dimensional grid modeling are combined, geometrical characteristics of static obstacles and motion parameters of dynamic obstacles are restored, and the collision risk caused by environmental perception errors of the mechanical arm is effectively avoided.

The invention provides an agent path planning method based on improved RRT and related equipment, and the method comprises the steps: 1, obtaining a current position and a target position of an agent, and constructing a search tree; 2, in the expansion process, sampling is carried out based on a dynamic weight target bias sampling strategy; extending a distance of a dynamic step length from the extension starting point to the extension direction based on a dynamic step length adjustment strategy to obtain a new node, and performing kinematics constraint detection and collision detection on the new node; 3, obtaining an initial path through path backtracking; step 4, when a dynamic obstacle is detected, re-planning a path section conflicting with the dynamic obstacle in the initial path based on an incremental re-planning strategy to obtain a final path for indicating the intelligent agent to move; otherwise, taking the initial path as a final path. According to the invention, the accuracy of path planning can be improved.

The invention discloses an unmanned vehicle intelligent obstacle avoidance method and system based on a multi-modal sensor, and relates to the field of automatic driving and intelligent traffic. According to the technical key points of the invention, data of a laser radar and a depth camera are aligned through a space-time registration module; the laser radar point cloud is preprocessed, obstacle point cloud is segmented, fusion clustering is carried out in combination with three-dimensional semantic seed points output by a depth camera, and an obstacle cluster with a semantic tag is generated; and calculating a three-dimensional directed bounding box (OBB) of an obstacle, dynamically adjusting a safety distance according to the speed of the vehicle, expanding the OBB of the vehicle, and carrying out collision detection by adopting a separation axis theorem. When a collision risk is detected, the system sequentially triggers a first-level early warning instruction and a second-level emergency braking instruction, and safe and efficient unmanned vehicle dynamic obstacle avoidance is achieved. According to the method, the advantages of multiple sensors are fully utilized, the obstacle detection precision and the response speed are effectively improved, and reliable technical support is provided for application of the unmanned driving technology in a complex road environment.

Systems and methods are described for obtaining accurate responses from large language models (LLMs) and chatbots, including for question and answering, exposition, and summarization. These systems and methods accomplish these objectives via use of noun phrase avoiding processes such as a noun phrase collision detection process, a query splitting process, and a topical splitting process as well as by use of formatted facts, formatted fact model correction interfaces (FF MCIs), bounded-scope deterministic (BSD) neural networks, processes and methods, and intelligent storage and retrieval (ISAR) systems and methods. These systems and methods avoid and bypass noun phrase collisions and correct for errors caused by noun phrase collisions so that hallucinations are eliminated from LLM responses.

The invention discloses a multi-fingered dexterous hand grabbing posture generation method and system based on CVAE and a Ball Query algorithm, and belongs to the technical field of robot grabbing control. The method comprises the following steps: (1) a data sampling step; (2) a data preprocessing step: carrying out standardization processing on the collected data, generating enhanced point cloud data and constructing a training data set; (3) a model training step: through a multi-scale feature extraction module, combining global semantics and local geometric features extracted by a Ball Query algorithm, generating a grabbing attitude by using a conditional variation auto-encoder, and optimizing model parameters through reconstruction loss and KL divergence; and (4) real-time deployment: integrating the trained model to a physical platform, screening candidate grabbing postures based on parallel collision detection, and realizing real-time grabbing control in combination with inverse kinematics verification. The problem that a traditional method is insufficient in generalization ability in complex object grabbing is effectively solved, and self-adaptive grabbing of unknown objects is achieved while grabbing stability is guaranteed.

A three-dimensional encasement reinforcement learning method based on multi-modal sensing distillation comprises the steps that a high-fidelity physical simulation environment is constructed, strategy optimization is achieved through a teacher-student double-stage framework, a teacher network fuses vision, article attributes and internal physical state privilege information, and a high-stability encasement strategy is trained through hierarchical action masks and domain randomization; a student network only depends on visual input, teacher action distribution and value functions are aligned through KL divergence and L2 loss, and knowledge distillation is achieved; after deployment, a motion mask is generated in combination with lightweight collision detection. A failure scene can be played back to a simulation environment through a playback mechanism, and a student network is continuously optimized by using a teacher strategy. According to the method, the space utilization rate is remarkably improved in simulation and reality, the turnover rate is reduced, complex physical constraints can be adapted without expensive sensors, and an efficient and robust solution is provided for industrial automation.

The invention provides a machine grabbing learning method fusing shape features, contact modeling and physical constraints, and the machine grabbing learning method is a robot grabbing posture learning method capable of differentiating. According to the method, a three-dimensional point cloud serves as input, a deep neural network based on sparse voxel convolution is adopted to extract multi-scale shape features, and potential grabbing points and attitude parameters thereof are predicted. The differential contact modeling is realized by constructing a local area of direction perception and extracting geometric features reflecting the contact relationship between the mechanical claw and the object. Furthermore, the collision probability of the grabbing posture is estimated based on an implicit neural network, and a differentiable collision detection module is constructed. Meanwhile, physical constraints such as friction closing, surface alignment and geometric symmetry are introduced to serve as regular terms to jointly optimize grabbing scores, collision risks and stability; according to the method, end-to-end training is supported, efficient, robust and deployable grabbing posture estimation is achieved, and the method is suitable for various automatic grabbing tasks such as industrial robots and service robots.

The invention discloses a mobile robot path planning method and system based on a circular trajectory control obstacle function, and aims to realize trajectory smooth control in an obstacle avoidance process. Firstly, an envelope circle is used for modeling an obstacle, an actual boundary is approached in a continuous and simple geometric form, and the calculation complexity of collision detection is reduced. Secondly, designing a control obstacle function based on a circular trajectory, constructing a derivable constraint in combination with a geometrical relationship between a tangent line and an obstacle boundary, and achieving trajectory smoothing, reducing speed fluctuation and prolonging the service life of an execution mechanism while guaranteeing a safe distance. For disturbance caused by factors such as ground unevenness, historical data are learned by using a Gaussian process, the disturbance trend is predicted in real time, and the motion stability is improved. And finally, through nonlinear optimization solution, generating control input with safety, continuity and disturbance adaptive capability. According to the method, geometric modeling, safety constraint, disturbance prediction and a dynamic adjustment mechanism are fused, and the safety and adaptability of robot path planning are remarkably enhanced.

The invention provides a cylinder processing cutter path adaptive optimization generation method, which belongs to the technical field of cylinder processing, and adopts fractal dimension analysis to identify surface geometric complexity and determine a minimum repetitive unit, and applies a four-color theorem to carry out non-intersection area division on the surface, so as to improve the processing precision of the cylinder processing cutter path. Multi-axis linkage machining parameters are initialized, global path planning and cutter axis vector optimization and collision detection are performed based on an improved Di jkstra algorithm, path adaptive adjustment is performed according to local curvature change of a minimum fractal matrix, a game model is established, and coevolution of a double-layer game model is realized by adopting a cross optimization rule. And finally, an optimized tool path NC code containing the position coordinates, the attitude angle and the feeding speed is generated, and the technical problem that the machining efficiency and the surface quality are difficult to consider in tool path planning under the complex geometrical characteristics of the cylinder surface is solved.

The invention relates to the field of mobile robot motion planning, and discloses a quadruped robot motion planning method and system based on height self-adaption. The method comprises the following steps: constructing 3D and 2D local obstacle maps according to coordinate transformation, and calculating a grid vertical distance through ray casting to generate a 2.5 D local obstacle map; determining an initial control sampling space by combining speed, acceleration and height constraints; discretizing a sampling space and simulation time, deducing a control value sequence and recursively calculating a local trajectory; then calculating a shaft alignment bounding box in a directed bounding box of the robot, screening grid units in the shaft alignment bounding box, comparing the height of a machine body with the allowable height of grids, and rejecting collision trajectories to obtain a feasible trajectory set; and finally selecting an optimal track by the evaluation function, extracting a control value and sending the control value to hardware for execution. According to the method, the 2.5 D local obstacle map containing height information is constructed, the four-dimensional control sampling space is designed, and the robot height self-adaptive adjustment is realized in combination with the collision detection and evaluation function.

The invention discloses a collision identification method and system for a window cleaning robot, and relates to the technical field of robot environment perception, and the method comprises the steps: carrying out the time difference compensation and space synchronization based on a glass material type and collision detection parameters through a distributed perception collaborative analysis method, and generating a three-dimensional space vibration information matrix; enhancing the three-dimensional space vibration information matrix through a dynamic wavelet packet decomposition algorithm and a sliding window analysis method according to the type of a glass material; according to the enhanced three-dimensional space vibration information matrix, collision coordinates are calculated through a time difference positioning algorithm, the characteristic frequency of the vibration waveform is analyzed, and collision characteristic information is obtained; based on the collision feature information, combining safety parameters of glass material types, adopting a hybrid decision algorithm to analyze the collision feature information, and generating and executing a robot obstacle avoidance instruction; according to the method, the high-resolution three-dimensional vibration field is constructed through space-time collaborative modeling, and the collision coordinate calculation precision and the anti-interference capability are improved.

The invention provides a three-dimensional space collision avoidance system based on multiple tower cranes, and relates to the technical field of port hoisting operation, a three-dimensional voxel grid VoxelGrid space map is constructed, and the three-dimensional space collision avoidance system comprises a first horizontal coverage map, a second descending coverage map, a third fan-shaped area occupation map, a fourth track occupation area map and a fifth interference area map; the system can identify the space overlapping area of the tower crane operation path at the voxel level, breaks through the limitation of the traditional path planning method on the dimension and precision of space modeling, and enhances the fine granularity and accuracy of collision detection. A dynamic acquisition module and a dynamic interference module are introduced, dynamic disturbance caused by factors such as stormy waves and tides on a ship in a port operation environment can be sensed in real time, the deviation trend of the tail end of the lifting appliance is predicted, target correction and track compensation are carried out before track execution, and the situation that the lifting appliance is accidentally touched or falls off due to track drifting is effectively promoted to be reduced.

The invention provides a three-dimensional object collision detection parcel grid generation method and system, a storage medium and a program product, and the method comprises the steps: extracting all vertex data of a first object according to a sub-Mesh data set after serialization of a three-dimensional model, and determining an initial center point; each vertex of each piece of sub Mesh data is converted from a local coordinate system to a global world coordinate system, and a point cloud data set is generated; constructing an initial convex hull by adopting a matrix hierarchical extremum retrieval algorithm, and generating a final convex hull by adopting a conflict graph and incremental convex hull algorithm; initializing a radius coverage model by taking the geometric center of the model as a sphere center; performing subdivision according to longitude and latitude to generate a hierarchical ball grid; emitting rays, correcting vertex positions layer by layer, approaching the surface of the model, and outputting a collider network. Through the overall point cloud set, the incremental convex hull algorithm and the spherical grid contraction technology, unified topological wrapping of the multiple three-dimensional objects is achieved, the visual fidelity and the collision precision are ensured, and resource consumption is reduced.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes such as automatic driving and intelligent transportation, financial science and technology, medical health and the like, and discloses a three-dimensional scene event analysis method, device, equipment and medium. Generating a three-dimensional bounding box for a dynamic target in each frame of multi-view image data, fusing the information of the three-dimensional bounding box with the features of the semantic graph, generating a refined three-dimensional voxel semantic graph, aggregating multiple frames of refined three-dimensional voxel semantic graphs, constructing a motion track of the dynamic target, and detecting a space overlapping event based on the motion track. And generating a collision event description based on the relative motion state, and generating an attribution analysis result of the event. According to the method, the accuracy and automation level of collision detection and responsibility analysis in a dynamic scene are improved by fusing the multi-frame image information and the three-dimensional voxel data and combining the target position and the motion state.

The invention discloses a building collision detection method based on BIM cooperative multi-agent reinforcement learning, and belongs to the technical field of building engineering. The method comprises the following steps: firstly, acquiring geometric data and pipeline distribution information from a preset building information model, and determining model block granularity through analysis to obtain a sub-model set; then, a multi-agent system is adopted to distribute monitoring responsibilities, edge geometrical shape data and physical connection point locations are extracted, potential conflict areas with overlapped boundaries are judged, and a collision probability distribution diagram is generated; exchanging real-time data by setting an information synchronization period between intelligent agents to obtain local conflict distribution details, introducing real-time change data to train the intelligent agents by adopting a reinforcement learning algorithm, and updating a monitoring priority and a cooperative action sequence; and if the conflict points which are not solved still exist, local collision detection is carried out by adopting a distributed computing framework, and finally, efficient and accurate building information model collision detection is realized by adaptively adjusting sub-model division and task allocation rules.

The invention discloses an automatic obstacle avoidance method and system for an unmanned ship based on Beidou artificial intelligence, and relates to the field of computer vision, and the method comprises the steps: collecting a visible light image and a near-infrared image of a water surface environment, and carrying out the reflection suppression processing; performing feature fusion on the image to generate an environment perception graph; obstacle features are extracted through a cascade cavity convolution structure, and an obstacle pixel-level semantic segmentation result is obtained; constructing an obstacle motion prediction model, and obtaining a predicted obstacle motion track; acquiring motion track data of the unmanned ship based on Beidou positioning and electronic chart data; establishing a space-time collision detection model, calculating collision parameters, and generating a collision data set; an obstacle avoidance path is generated through a dynamic situation field method, and a rudder angle instruction and a propeller rotating speed instruction are obtained. The unmanned ship has the advantages that accurate sensing, motion prediction and intelligent obstacle avoidance of the unmanned ship on water surface obstacles are realized through Beidou positioning and technology, and the sailing safety and intelligent level are improved.

The invention discloses a virtual assembly and anti-collision simulation method for a double-swing-head five-axis numerical control machine tool, and the method comprises the following steps: S1, building a geometric model of the double-swing-head five-axis numerical control machine tool based on a machine tool kinematic model, and the geometric model comprises parameterized three-dimensional models of a main shaft part, a swing head mechanism, a workbench and a machine tool body; s2, according to the constraint relation of the kinematic chains of the machine tool, an assembly hierarchical relation is established in the virtual environment, and virtual assembly of all parts is achieved through dynamic constraint driving; s3, on the basis of a machine tool kinematics forward and inverse solution algorithm, simulating the motion trail of the double swing heads in the five-axis linkage process; s4, through a real-time collision detection algorithm, detecting interference conditions among the cutter, the swing head, the workpiece, the clamp and the machine tool body, and generating an anti-collision early warning signal; and S5, optimizing a motion path of the machine tool in combination with a simulation result, and outputting a collision-free machining code. The assembly precision can be improved, and the physical debugging cost can be reduced.

The invention relates to the technical field of building steel structure construction, and provides a closed-loop control method for segmented construction of a special-shaped curtain wall steel structure, and the method comprises the steps: S101, carrying out the three-dimensional scanning of a to-be-installed component and a supporting jig frame, building a coordinate system through a positioning target, and obtaining point cloud data; step S102, extracting component splicing key points, matching point cloud with a design model, performing collision detection and generating pre-splicing positioning data; s103, converting the positioning data into a visual instruction, and guiding the component to be accurately in place in combination with the real-time hoisting data; s104, after installation, a seam area is scanned, and a new entering component is reinspected; s105, the step S102 to the step S104 are repeated until all the components are installed; and S106, integrally scanning to generate a completion model, comparing the completion model with the design model, and outputting a construction quality acceptance report. According to the technical scheme, the construction precision, efficiency and construction quality of the special-shaped curtain wall steel structure are collaboratively optimized by constructing a closed-loop control process.

The invention discloses an aircraft control simulation platform construction method, and the method comprises the steps: constructing a terrain multi-dimensional model, fusing a cross-scale meteorological particle system, and dynamically obtaining an environment disturbance factor; a space obstacle dynamic database is generated in combination with millimeter-level space positioning data, and virtual collision detection is carried out; performing eye movement-gesture collaborative decision, and identifying a control instruction with high confidence; the six-degree-of-freedom kinetic equation is coupled with an unreal engine interface in real time, and a high-fidelity instrument is driven to feed back; generating a training defect thermodynamic diagram, triggering an adaptive optimization strategy and iterating platform parameters; the invention provides a solution integrating multi-modal high-precision data acquisition, dynamic environment-physical rule coupling, intelligent decision and closed-loop optimization so as to realize aircraft control simulation training with high immersion degree, high fidelity and high adaptability.

The invention provides a real-time monitoring system and method for sorting of a digital twin-driven Delta mechanical arm. According to the system, firstly, a physical experiment platform is built, and a virtual model is built through SolidWorks, 3DMAX and Unity 3D. A communication framework is designed, and wireless communication of the PLC, the sensor, the mechanical arm, the conveying belt and the digital twin system is achieved. Positive and inverse kinematics analysis and Cartesian space trajectory planning are carried out on a Delta robot, and a related algorithm is packaged into an MATLAB function and compiled into a DLL library for C # calling. The system collects joint angle data in real time based on a rotary encoder, transmits the joint angle data to a digital twin system through an OPC UA protocol, and drives a virtual mechanical arm model after resolving, so that high-precision and low-delay virtual-real synchronous pose reconstruction is realized. In addition, the system also develops a collision detection function based on a bounding box algorithm, and performs visual detection by using camera calibration and a YOLO model.

The invention relates to a mechanical arm visual servo control method and system based on deep reinforcement learning. Firstly, real-time state observation and collision detection are conducted on the mechanical arm, and an observation result is obtained; then, a deep reinforcement learning controller is built by means of a maximum entropy deep reinforcement learning SAC algorithm network architecture, and an action space, a state space and a reward function are built according to an observation result and a target position; and then, outputting a first action representation based on the construction content, optimizing a controller according to a calculation result of the first action representation and the reward function, smoothly optimizing the first action representation by using a real-time track interpolation strategy to obtain a second action representation, and controlling the mechanical arm joint according to the second action representation. And the process is continuously repeated until the mechanical arm successfully arrives at the target position. Compared with the prior art, the method has the advantages of high control real-time performance, large application range and the like.

The present invention relates to the technical field of path planning, and provides a deep reinforcement learning-based path exploration parameter optimization system. The system comprises: a variable parameter path planning module, configured to perform node exploration based on a deep reinforcement learning network, conduct collision detection on child nodes in a child node set, calculate cost values for all child nodes, and finally generate a loading and parking path using a Reeds-Shepp curve; an environmental state space modeling module, configured to perform regional division of obstacles around a current node and conduct environmental state space modeling; and a deep learning parameter optimization module, configured to construct a deep learning network to compute an optimal step size and an optimal steering angle, build a reward function to optimize the deep learning network, and simultaneously execute a training process of the deep learning network.

The invention discloses a task-oriented grabbing method and system based on cross-level constraint reasoning, and belongs to the technical field of robot grabbing control. The method comprises the following steps: aligning a text instruction with an input image through a mask alignment module, generating a target area mask by utilizing SAM-Clip, and generating a target point cloud by taking the target area mask as spatial prior; and further analyzing an internal physical structure of the target by using VLM, guiding to generate a 6-DOF grabbing attitude, performing collision detection and quality scoring by combining high-level function and bottom-level geometric prior, and outputting an optimal grabbing attitude. According to the method, the problems that task understanding and scene perception are disjointed, and the grabbing posture lacks constraint are solved, and the grabbing success rate and the intelligent level of the robot in the open environment are remarkably improved.

The invention discloses a pediatric nursing practice teaching VR model generation method and system based on a training video, and relates to the technical field of artificial intelligence and virtual reality, and the method comprises the steps: collecting a pediatric nursing operation video for preprocessing, extracting a key motion frame, separating medical staff, a child patient and a medical tool, converting the key motion frame into an editable 3D model, and carrying out the recognition of the editable 3D model; acquiring multi-modal data corresponding to the pediatric nursing operation, matching the converted 3D actions with semantics through a multi-modal large model, and optimizing 3D conversion; key contact points are marked in the 3D action sequence to carry out collision detection point burying, an operation process is defined to generate a 3D animation sequence with interaction logic to construct a VR teaching scene, the reasonability of the operation logic is verified through user testing, and scene correction is carried out. Through video analysis, 3D element replacement and animation point burying technologies, intelligent generation of pediatric nursing VR teaching resources is realized, and a more efficient and more immersive solution is provided for nursing education.

The invention provides a large-diameter long-pipe-section pipe pile hoisting path planning method and device, and relates to the technical field of cranes. According to the method, motion state parameters of pipe sections and pipe piles are obtained in real time, and in combination with three-dimensional environment sensing data of a hoisting area, a space operation diagram containing obstacles and a safety buffer area is constructed; generating a plurality of groups of candidate paths combined by speeds and steering angles in the fan-shaped region according to a set angle interval by taking the current motion direction as a reference; removing paths conflicting with obstacles through collision detection, and reserving an effective path set; based on multi-dimensional feature analysis of a security item index, a smoothness index and a progress index, a scoring function is constructed by presetting a weight, and an optimal path is dynamically selected; path planning is carried out through a fixed time interval until the pipe pile stably arrives at a target point; through real-time environment modeling and dynamic path planning, the problems of obstacle avoidance and efficiency of large-diameter pipe pile hoisting in a complex scene are solved, and hoisting safety and operation accuracy are remarkably improved.

The invention provides a complex grid cutting method based on an autonomous cutting algorithm. The complex grid cutting method comprises the following steps: preprocessing a cutting grid and a to-be-cut grid body; performing collision detection on the preprocessed cutting grid and the grid body to be cut; carrying out intersection calculation on the intersected triangles based on a detection result to obtain an intersection line; respectively embedding the obtained intersecting lines into the cutting grid and the grid body to be cut; carrying out topological separation on the to-be-cut grid body based on intersecting lines, and marking a part located in the positive direction of the cut grid as a first sub-grid; topological separation is carried out on the cutting network based on intersecting lines, and the part located in the to-be-cut grid body is marked as a second sub-grid; and sewing the first sub-grid and the second sub-grid to serve as a final cut sub-grid body. The method solves the problem of low success rate of Boolean operation of complex grids.

The invention belongs to the technical field of mobile mechanical arm grabbing, and provides a mobile mechanical arm grabbing control method and system based on machine vision, and the method comprises the steps: collecting an image in the moving process of a mobile mechanical arm according to a preset grabbing path, recognizing a dynamic obstacle in the image through an RT-DETR algorithm, and generating a prediction track of the dynamic obstacle, collision detection is conducted on a preset grabbing path of the movable mechanical arm, an obstacle avoidance track switching instruction is generated by combining the current time point of the mechanical arm with the predicted collision time point of the dynamic obstacle, trajectory tracking is conducted on the obstacle avoidance track of the mechanical arm, a model reference self-adaptive control algorithm is adopted, a reference model is established, and obstacle avoidance is achieved. According to the method, mechanical arm vibration caused by obstacle avoidance track switching is effectively restrained, interference of vibration on grabbing stability is reduced, and the operation precision and reliability of the mechanical arm in a dynamic obstacle avoidance scene are remarkably improved.

The invention discloses an obstacle avoidance motion planning method and system for redundant drive of an intelligent process of a welding robot and computer equipment. According to the method, firstly, a motion path of a robot is planned based on welding process redundancy, then a sampling strategy that the position and posture of a welding gun are separated is adopted, then the sampled posture is mapped to a planar two-dimensional space, and the relevance between an obstacle collision boundary and the obstacle avoidance posture of the welding gun is constructed; an incremental extreme gradient lifting model is adopted to carry out probability modeling on the collision boundary and obstacle avoidance attitude relevance, nearest neighbor search and new planning node expansion are carried out based on target cost to obtain new planning nodes, collision detection is carried out on the new planning nodes, and the new planning nodes passing the collision detection are used for incremental learning of an obstacle avoidance model; and finally, an optimization strategy of spherical linear interpolation and cosine slow motion time mapping is adopted to ensure continuity and smoothness of a motion planning path trace. The problem that in the prior art, it is difficult to quickly calculate and generate a collision-free welding track is solved.

The invention discloses a vehicle collision detection, description and early warning system and method based on a driving video, and belongs to the technical field of artificial intelligence and intelligent traffic safety. According to the system, on the basis of a vision-language model, vehicle-mounted videos such as an automobile data recorder are automatically analyzed, and detection, severity grading, natural language description generation and real-time early warning of vehicle collision events are achieved. The system extracts high-dimensional semantic features of video frames through a CLIP model, focuses key information through an attention weighting network, inputs the key information into a deep classification network composed of a plurality of full connection layers, a batch normalization layer and an activation function, and outputs a multi-level accident severity classification result. Integrating target detection and environment information, and generating a structured accident description text by using a fine tuning BART model; a sliding window and time sequence modeling mechanism is adopted, and recognition and early warning of the pre-collision state are achieved. The method effectively solves the technical problems of lack of semantic understanding, inaccurate severity judgment, incapability of early warning and the like of a traditional method, has high accuracy, high interpretability and real-time response capability, and is suitable for intelligent driving assistance and traffic safety monitoring scenes.