Multi-task processing method and device for laser point cloud data and moving tool
Through the multi-task processing method of laser point cloud data, the multi-task model is generated using pre-training and joint training strategies, the problem of high computing power consumption in autonomous driving is solved, and real-time and efficient on the on-board platform is achieved The processing improves the performance of autonomous driving environment perception.
Patent Information
- Application Number
- CN202310546324.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-15
- Publication Date
- 2024-11-15
AI Technical Summary
In the field of autonomous driving, the large amount of laser point cloud data and high computing power consumption leads to challenges in real-time and accuracy of vehicle embedded platforms, especially in complex environments in urban scenarios, where the operation of multiple independent models is limited Computing power is a difficult problem.
The multi-task processing method of laser point cloud data is adopted to generate detection and pass area identification input data by processing the collected sample data, and a pre-training and joint training strategy is used to generate multi-task models to realize multi-task processing of a single model. Reduce memory and computing power consumption.
Real-time and efficient operation on the vehicle-mounted embedded platform is achieved, and the performance of laser point cloud data processing is improved, which not only reflects the connection between tasks but also reflects the differences, saving the trouble of multi-task repetitive sampling.
Smart Images

Figure CN118968459A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a multi-task processing method, device and mobile tool for laser point cloud data. Background Art
[0002] Autonomous driving is a systematic project, in which environmental perception is a very important link in the whole system; the results of environmental perception directly affect the quality and effect of back-end links such as positioning and regulation and control. Environmental perception mainly uses various types of sensors to quantitatively describe the surrounding environment and output it in different data forms. At present, there are two main types of environmental perception: one is a visual perception solution that is mainly based on visual images and supplemented by radar (lidar, millimeter-wave radar, etc.); the other is a laser perception solution that is mainly based on lidar and supplemented by visual sensors such as cameras. Lidar can accurately perceive the three-dimensional information of the surrounding environment with its high detection accuracy; and lidar can better identify the specific outline and distance of the target. Therefore, laser perception solutions are widely used in the field of autonomous driving.
[0003] However, the data volume of laser point cloud is large, and the computing power consumption is large during operation. However, the computing power of the vehicle embedded platform is limited, so the real-time requirements for laser point cloud related tasks are high. Therefore, achieving both real-time and accuracy becomes the key and challenge of this task.
[0004] Usually, using specialized models and algorithms for independent tasks is the most accurate and direct way to solve problems. Due to the complexity of autonomous driving operation scenarios, especially urban scenarios, multiple models are usually required to solve the problem. The independent operation of multiple independent models is a challenge for the limited computing power of the vehicle embedded platform. Summary of the invention
[0005] The purpose of the present invention is to provide a multi-task processing method for laser point cloud data in view of the defects of the prior art. Different data processing is performed for different tasks, and different model training strategies are adopted to realize multi-task processing of a single model.
[0006] To achieve the above object, the present invention provides a multi-task processing method for laser point cloud data in a first aspect, comprising:
[0007] Processing the collected sample laser point cloud data to generate sample detection input data and passable area identification input data; the sample detection input data includes sample categories;
[0008] Pre-training a preset model according to the sample detection input data to obtain a pre-trained model;
[0009] The sample detection input data and the traversable area recognition input data are input into the pre-trained model for joint training to obtain a multi-task model, where the multi-tasks include a target detection task and a traversable area recognition task.
[0010] In a second aspect, the present invention provides a multi-task processing device for laser point cloud data, comprising:
[0011] An input data generation module is used to process the collected sample laser point cloud data to generate sample detection input data and passable area identification input data; the sample detection input data includes sample categories;
[0012] A model pre-training module, used to pre-train a preset model according to the sample detection input data to obtain a pre-trained model;
[0013] The multi-task model generation module is used to input the sample detection input data and the traversable area recognition input data into the pre-trained model for joint training to obtain a multi-task model, wherein the multi-task includes a target detection task and a traversable area recognition task.
[0014] According to a third aspect of the present invention, a computer-readable storage medium is provided, comprising a program or an instruction. When the program or the instruction is executed on a computer, the multi-tasking method as described in any one of the first aspects is implemented.
[0015] According to a fourth aspect of the present invention, a computer program product is provided. When the computer program product is run on a computer, the computer is enabled to execute the multi-tasking method as described in any one of the first aspects.
[0016] According to a fifth aspect of the present invention, a chip system is provided, comprising a processor coupled to a memory, wherein the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the multi-tasking method described in any one of the first aspects is implemented.
[0017] A sixth aspect of the present invention provides a computer system, comprising a memory, and one or more processors communicatively connected to the memory;
[0018] The memory stores instructions that can be executed by the one or more processors, and the instructions are executed by the one or more processors to enable the one or more processors to implement the multi-tasking method as described in any one of the first aspects.
[0019] A seventh aspect of the present invention provides a mobile tool, comprising the computer system described in the sixth aspect.
[0020] The embodiment of the present invention provides a multi-task processing method for laser point cloud data. First, by processing the laser point cloud data accordingly according to different tasks, sample detection input data and passable area identification input data are generated, which saves the trouble of repeated sampling of multiple tasks and also makes the relationship between each task more in line with objective facts. Then, a training strategy of pre-training and joint training is adopted to train the multi-task model. Its running memory and computing power are greatly reduced, and multi-task processing of a single model is realized, which can not only reflect the connection between tasks, but also reflect the differences, and can well improve the operating performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 One of the flow charts of the multi-task processing method of laser point cloud data provided by an embodiment of the present invention;
[0022] Figure 2 Flow chart 2 of the multi-task processing method of laser point cloud data provided by an embodiment of the present invention;
[0023] Figure 3 Flowchart 3 of the multi-task processing method for laser point cloud data provided by an embodiment of the present invention;
[0024] Figure 4 A block diagram of a multi-task processing method for laser point cloud data provided by an embodiment of the present invention;
[0025] Figure 5 A module structure diagram of a multi-task processing device for laser point cloud data provided by an embodiment of the present invention;
[0026] Figure 6 A schematic diagram of the structure of a computer program product provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0027] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0028] A multi-task processing method for laser point cloud data provided by an embodiment of the present invention can be run in real time and efficiently in a vehicle-mounted embedded platform.
[0029] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments.
[0030] The embodiment of the present invention provides a multi-task processing method for laser point cloud data, which mainly includes the following steps: Figure 1The steps shown in:
[0031] Step 110, processing the collected sample laser point cloud data to generate sample detection data and sample passable area data.
[0032] The multi-tasks of the present application specifically include target detection tasks and traversable area recognition tasks. The sample detection data includes sample categories and sample quantities, wherein the sample quantity must meet a preset quantity threshold.
[0033] The step 110 can be specifically performed as follows: Figure 2 The substeps shown implement:
[0034] Step 1101 , annotate sample laser point cloud data according to preset rules to obtain annotated sample detection data and annotated sample passable area data.
[0035] Here, the data processing of the target detection task and the traversable area recognition task are explained respectively.
[0036] For data annotation of target detection tasks, point cloud features can be extracted through point cloud column related algorithms, so as to annotate sample laser point cloud data with target location information, target size, target orientation, and target category, and obtain annotated sample detection data. The target location information is annotated in the form of rectangular frame coordinates. The target size is annotated in the form of length, width, and height. Target categories can specifically include people and cars.
[0037] For the data annotation of the passable area identification task, the sample laser point cloud data is mainly processed into two dimensions, and multiple frames of continuous sample laser point cloud data are spliced to obtain the labeled sample passable area data.
[0038] Specifically, the 3D laser point cloud data is projected onto a 2D image from a bird's eye view (BEV) perspective to obtain a 2D feature map. In order to avoid interference to the annotation caused by too many point cloud points, as a preferred solution, the present application imposes a threshold limit on the height of the 3D laser point cloud data so that points whose height exceeds the threshold are not projected. As an example but not a limitation, the threshold may be 2 meters. Splicing multiple consecutive frames of sample laser point cloud data can improve the connectivity of the traffic area and the learning of complex traffic areas during the model learning process.
[0039] Step 1102 , preprocessing the labeled sample detection data and the labeled sample passable area data to generate sample detection input data and passable area input data.
[0040] Specifically, preprocessing can be understood as the process of converting the labeled sample detection data and the labeled passable area data into a data format that can be input into the model, and then enhancing it.
[0041] More specifically, first, the task category label corresponding to each sample is determined based on the labeled sample detection data and the labeled sample passable area data;
[0042] Among them, the labeled sample detection data and the labeled sample passable area data include task labels and the number of points of the labeled objects.
[0043] Secondly, the sample laser point cloud data are classified according to the task category labels to determine the sample set corresponding to each task.
[0044] Specifically, according to the task category label, filter the annotated samples required for this task. For example, according to the task category that needs to be trained, each sample has at least a preset number of point cloud points, and the number of task categories, determine how many annotated objects are set for each task category to form a training sample, and then rotate the training samples and filter the data range, etc., to obtain the sample set corresponding to each task. Among them, the preset number can be specifically 10. The target detection task is a subset of the traversable area recognition task. In this way, it can be ensured that the effect of the target detection task is equivalent to that of the proprietary detection model. The operation performance is improved through structural changes to the model and the previous real-time measures.
[0045] In a preferred solution, before executing step 110, the process further includes: Figure 3 Steps shown:
[0046] Step S1, determining the backbone network for the traversable area recognition task.
[0047] Specifically, this application needs to take into account both the target detection task of point cloud data and the task of traversable area identification. Therefore, the backbone network of the traversable area identification task is first determined. In this example, the backbone network draws on the main framework of STDC and makes appropriate deletions. More specifically, the ConvX2 stage of the original STDC is deleted; the Stage3, Stage4, and Stage5 stages of the original STDC are retained; in order to take into account the effect of the detection task, the activation function of the GlobalPool stage of the original STDC is changed to Sigmoid.
[0048] Step S2, determining the convolution combination for the target detection task based on the backbone network for the traversable area recognition task.
[0049] Specifically, STDC is mainly used for segmentation tasks. This application also requires target detection tasks. Therefore, based on the backbone network of the traversable area recognition task, some convolution combinations with good detection effects are appropriately added to achieve a balanced effect of multi-task processing.
[0050] Specifically, in the first step, the encoding process of the target detection task is the same as the combination of ConvX1, Stage3, Stage4, and Stage5 of STDC; in the second step, the decoding process of the target detection task is based on the output of the four stages in the first step. The features of the transposed convolution (ConvTranspose) of each branch in the second step are expanded (concat) and output to the prediction head (detction head) of the detection branch.
[0051] Step 120: pre-train the preset model according to the sample detection input data to obtain a pre-trained model.
[0052] Specifically, since the amount of detection data is relatively large and the convergence speed of the detection branch of the model is slow, pre-training is required to match the characteristics of the detection branch. According to the convolution combination of the target detection task, the feature map is extracted, and the peak value of the heat map is calculated to determine the center of the sample category, and then determine the detection box of the sample category.
[0053] In order to perform targeted regression for different categories, this application specifically uses an anchor-free center heatmap detection head as the detection head. First, by calculating the peak value of the heatmap, regress the center point of the detection box of each sample category, and then regress each edge of the detection box based on the center point and the extracted feature map, which can greatly improve the accuracy of the detection box.
[0054] Step 130: Input the sample detection input data and the sample traversable area recognition input data into the pre-trained model for joint training to obtain a multi-task model.
[0055] Specifically, the convergence speed of the traversable area recognition branch is faster. During joint training, the model can learn the relationship between traversable areas and target detection, reducing the mutual influence between the two.
[0056] To be more specific, the passable area recognition head is divided into two parts for training, detail feature prediction (Detail Head) and segmentation feature prediction (Seg Head) to improve learning ability. The structures of these two heads are basically the same, both consisting of 3*3Conv (convolution), BN (normalization) and ReLU (activation function). However, their roles in the model are not exactly the same. Detail Head learns more details from the feature map of the intermediate process in the intermediate stage of the backbone network, and is mainly used in the training stage; SegHead is applied to the model output stage, and different categories of feature representation are performed for the categories learned by the network. In this application, there are only two categories for passable areas, passable and impassable, and the confidence of different categories will be calculated as the final model output.
[0057] Furthermore, in order to enable the model to be calculated in real time in the vehicle embedded system and maintain good robustness, the present application also includes:
[0058] Firstly, an in-vehicle multi-task model is generated according to the multi-task model and the hardware parameters of the in-vehicle system.
[0059] Here, the trained model is mainly used to analyze and evaluate the complexity of operators in the model and the computational effect of operator combinations. The time-consuming and high-complexity operators are combined with the hardware parameters of the vehicle system and replaced with operators and combinations in an equivalent way, achieving equivalent functions while achieving time improvements and stable effects.
[0060] Secondly, the laser point cloud data collected by the vehicle-mounted lidar is processed to generate detection data and passable area data.
[0061] Specifically, the process of processing point cloud data here is the same as the data processing in the above model training process, which will not be repeated here.
[0062] Finally, the detection data and passable area data are input into the on-board multi-task model for processing to generate detection results and passable area results.
[0063] Figure 4 A block diagram of a multi-task processing method for laser point cloud data provided by an embodiment of the present invention, combined with Figure 4 This method is briefly introduced further.
[0064] First, the laser point cloud data is annotated, mainly for the joint annotation of detection data and passable area data. According to the annotated passable area data, the detailed features of the passable area are predicted. The annotated laser point cloud data is prepared before the model is input, that is, data preprocessing, which includes data enhancement processing, and then the prepared data is input into the model to pre-train the model, that is, the pre-training is only for training the target detection data. After that, the predicted passable area detailed features are input into the pre-trained model for joint training, that is, the target detection data and the passable area data are trained at the same time. Specifically, two steps are trained as a group, one step for training target detection data, and one step for training passable area data, so as to obtain a multi-task model, and at the same time realize the prediction of detection features and passable area segmentation features, and then after model post-processing, the target detection results and passable area results are obtained.
[0065] The embodiment of the present invention provides a multi-task processing method for laser point cloud data. First, by processing the laser point cloud data accordingly according to different tasks, sample detection input data and passable area identification input data are generated, which saves the trouble of repeated sampling of multiple tasks and also makes the relationship between each task more in line with objective facts. Then, a training strategy of pre-training and joint training is adopted to train the multi-task model. Its running memory and computing power are greatly reduced, and multi-task processing of a single model is realized, which can not only reflect the connection between tasks, but also reflect the differences, and can well improve the operating performance.
[0066] Embodiment 2
[0067] Figure 5 This is a module structure diagram of a multi-task processing device for laser point cloud data provided in the second embodiment of the present invention. The device may be a device capable of implementing the method provided in the first embodiment of the present application, such as a central processing unit or server of an unmanned vehicle. The multi-task processing device for laser point cloud data includes:
[0068] The input data generation module 10 is used to process the collected sample laser point cloud data to generate sample detection input data and passable area identification input data; the sample detection input data includes sample categories;
[0069] The model pre-training module 20 is used to pre-train the preset model according to the sample detection input data to obtain a pre-trained model;
[0070] The multi-task model generation module 30 is used to input the sample detection input data and the traversable area recognition input data into the pre-trained model for joint training to obtain a multi-task model, where the multi-tasks include the target detection task and the traversable area recognition task.
[0071] A multi-task processing device for laser point cloud data provided in the second embodiment of the present invention can execute the method steps in the above method embodiment, the input data generation module 10 implements step 110, the model pre-training module 20 implements step 120, and the multi-task model generation module 30 implements step 130. The specific implementation principle and technical effect are similar and will not be repeated here.
[0072] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the input data generation module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-mentioned determination module. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.
[0073] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more microprocessors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).
[0074] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The above computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.
[0075] Embodiment 3
[0076] Embodiment 3 of the present invention provides a computer-readable storage medium, including a program or an instruction. When the program or the instruction is executed on a computer, the multi-tasking method as described in any one of the above embodiments 1 is implemented.
[0077] Embodiment 4
[0078] In some optional embodiments, any of the multi-tasking methods provided in Embodiment 1 may be implemented as computer program instructions encoded in a machine-readable format on a computer-readable storage medium or encoded on other non-transitory media or products. Figure 6 Schematically illustrates a conceptual partial view of an example computer program product arranged according to at least some of the embodiments presented herein, the example computer program product comprising a computer program for executing a computer process on a computing device. In one embodiment, the example computer program product is provided using a signal-bearing medium. The signal-bearing medium may include one or more program instructions, which when executed by one or more processors may provide any of the multi-tasking methods provided in the first embodiment above to obtain a trained multi-tasking model. For example, Figure 1One or more features of steps 110 to 130 shown may be undertaken by one or more instructions associated with a signal-bearing medium. In some examples, the signal-bearing medium may include a computer-readable medium, such as, but not limited to, a hard drive, a compact disk (CD), a digital video disk (DVD), a digital tape, a memory, a ROM, or a RAM, etc. In some embodiments, the signal-bearing medium may include a computer-recordable medium, such as, but not limited to, a memory, a read / write (R / W) CD, a R / W DVD, etc. In some embodiments, the signal-bearing medium may include a communication medium, such as, but not limited to, a digital and / or analog communication medium (e.g., a fiber optic cable, a waveguide, a wired communication link, a wireless communication link, etc.).
[0079] Embodiment 5
[0080] Embodiment 5 of the present invention provides a chip system, including a processor, the processor is coupled to a memory, the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the multi-tasking processing method described in any one of the above embodiments is implemented.
[0081] Embodiment 6
[0082] Embodiment 6 of the present invention provides a computer system, including a memory, and one or more processors communicatively connected to the memory;
[0083] The memory stores instructions that can be executed by the one or more processors, and the instructions are executed by the one or more processors to enable the one or more processors to implement the multi-tasking method as described in any one of the above-mentioned embodiments.
[0084] Embodiment 7
[0085] Embodiment 7 of the present invention provides a mobile tool, including the computer system provided by the aforementioned embodiment 6.
[0086] Among them, the mobile tool can be any movable tool, such as vehicles (such as floor scrubbers, vacuum cleaners, sweepers, logistics vehicles, passenger cars, buses, coaches, vans, trucks, trucks, trailers, trailers, cranes, excavators, bulldozers, road trains, sweepers, sprinkler trucks, garbage trucks, engineering vehicles, rescue vehicles, logistics carts, automatic guided vehicles (Automated Guided Vehicle, AGV)), motorcycles, bicycles, tricycles, carts, robots, sweepers, balance bikes, etc. The present application does not strictly limit the types of mobile tools, and they are not listed here in detail.
[0087] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0088] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0089] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A multi-task processing method for laser point cloud data, characterized in that: include: Process the collected sample laser point cloud data to generate sample detection input data and passable area recognition input data; The sample detection input data includes sample categories; Pre-training a preset model according to the sample detection input data to obtain a pre-trained model; The sample detection input data and the traversable area recognition input data are input into the pre-trained model for joint training to obtain a multi-task model, where the multi-tasks include a target detection task and a traversable area recognition task.
2. The multitasking method according to claim 1, characterized in that: The processing of the collected sample laser point cloud data to generate sample detection data and sample passable area data specifically includes: The sample laser point cloud data is labeled according to preset rules to obtain labeled sample detection data and labeled sample passable area data; The labeled sample detection data and the labeled sample traversable area data are preprocessed to generate sample detection input data and traversable area input data.
3. The multitasking method according to claim 2, characterized in that: The sample laser point cloud data is labeled according to a preset rule to obtain labeled sample detection data and labeled sample passable area data, specifically including: Annotating the sample laser point cloud data with target location information, target size, target orientation, and target category to obtain annotated sample detection data; The sample laser point cloud data is two-dimensionally processed, and multiple frames of continuous sample laser point cloud data are spliced to obtain labeled sample passable area data.
4. The multitasking method according to claim 1, characterized in that: Before the collected sample laser point cloud data is processed, the method further includes: Determining a backbone network for the traversable area identification task; According to the backbone network of the traversable area recognition task, a convolution combination of the target detection task is determined.
5. The multitasking method according to claim 4, characterized in that: The pre-training of a preset model according to the sample detection input data to obtain a pre-trained model specifically includes: According to the convolution combination of the target detection task, the feature map is extracted and the peak value of the heat map is calculated to determine the center of the sample category and further determine the detection frame of the sample category.
6. The multitasking method according to claim 1, characterized in that: Also includes: Generate an in-vehicle multi-task model according to the multi-task model and hardware parameters of the in-vehicle system; Process the laser point cloud data collected by the vehicle-mounted laser radar to generate detection data and passable area data; The detection data and the passable area data are input into the vehicle-mounted multi-task model for processing to generate detection results and passable area results.
7. A multi-task processing device for laser point cloud data, characterized in that: include: An input data generation module is used to process the collected sample laser point cloud data to generate sample detection input data and passable area recognition input data; The sample detection input data includes sample categories; A model pre-training module, used to pre-train a preset model according to the sample detection input data to obtain a pre-trained model; The multi-task model generation module is used to input the sample detection input data and the traversable area recognition input data into the pre-trained model for joint training to obtain a multi-task model, wherein the multi-task includes a target detection task and a traversable area recognition task.
8. A computer-readable storage medium, characterized in that: The method comprises a program or an instruction, and when the program or the instruction is executed on a computer, the task processing method according to any one of claims 1 to 6 is implemented.
9. A chip system, characterized in that: The invention comprises a processor coupled to a memory, wherein the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the multi-tasking method according to any one of claims 1 to 6 is implemented.
10. A computer system, characterized in that: comprising a memory, and one or more processors communicatively connected to the memory; The memory stores instructions that can be executed by the one or more processors, and the instructions are executed by the one or more processors to enable the one or more processors to implement the multi-tasking method according to any one of claims 1 to 6.
11. A mobile tool, characterized in that: A computer system comprising the method of claim 10.
Citation Information
Patent Citations
Deep learning-based point cloud three-dimensional object detection method
CN113095172A
Travelable area prediction method, model, device and vehicle
CN115147798A
Multi-task detection method and device, storage medium and terminal
CN115294539A
Road target detection method and device, electronic equipment and storage medium
CN115775366A
Multi-task semantic segmentation method and device and electronic equipment
CN115797895A
