A system and method for improving performance of an in-vehicle computing platform

Abstract

The application provides a system and method for improving the performance of a vehicle-mounted computing platform, the system comprising an information acquisition unit, a central processing unit and an image processor; the information acquisition unit is used to acquire image data and point cloud data during vehicle driving; the central processing unit is used to acquire the image data and the point cloud data, and send the image data and the point cloud data to the image processor after preprocessing; and according to the processing result of the image processor on the image data and the point cloud data, corresponding operations are performed on the vehicle; after the image processor acquires batch image data, an object detection algorithm is used to search for abnormal behaviors affecting vehicle operation in the smallest search range, feature points of the point cloud data are extracted, and calculation tasks in the vehicle driving process are performed. Based on the system, a method for improving the performance of the vehicle-mounted computing platform is also provided, and the central processing unit and the image processor are used for cooperative operation, so that the computing power of the vehicle-mounted computing platform is maximized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent driving, and particularly relates to a system and method for improving the performance of a vehicle-mounted computing platform. BACKGROUND

[0002] One of the purposes of the Hermes vehicle-mounted computing platform project is to make resource and technology reserves for L4 unmanned vehicle-mounted computing, while meeting the needs of existing customers to achieve sales targets and lay the foundation for exploring the unmanned market. The Hermes vehicle-mounted computing platform will be applied to the EIS400 productization, providing technical support for the landing of the EIS400 in the low-speed unmanned vehicle-mounted computing market, while solving the problems of scattered traditional unmanned platform devices, high overall power consumption, and complex interconnection wire harnesses.

[0003] The existing computing task distribution scheme is that component computing is mainly based on the GPU of Orin, such as Figure 1 The flowchart of the prior art for distributing arbitrary tasks is given. For example, design AI model computing, vehicle-mounted camera, and laser radar data computing; and the Orin carrier board is equipped with a multi-core CPU and a high-performance GPU card, and the idle state of the CPU leads to resource waste. When designing AI model computing, vehicle-mounted camera, laser radar, and other components generate computing tasks, the tasks are distributed to the vehicle-mounted GPU processing module and the vehicle-mounted coprocessing module for processing. However, the prior art has obvious shortcomings, that is, the Orin carrier board is equipped with a multi-core CPU and a high-performance GPU card, and only the GPU is used for computing in the original design, which fails to achieve maximum computing power under limited physical resources; and there is no fault-tolerant mechanism to handle the GPU downtime, and once the GPU operation is down, it will cause a snowball effect of the Hermes vehicle-mounted computing platform. SUMMARY

[0004] To solve the above technical problems, the present application proposes a system and method for improving the performance of a vehicle-mounted computing platform, which uses a central processor and an image processor to solve the problem of not being able to fully utilize hardware performance.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] A system for improving the performance of a vehicle-mounted computing platform, comprising an information acquisition unit, a central processor, and an image processor;

[0007] The information acquisition unit is used to acquire image data and point cloud data during vehicle driving;

[0008] The central processor is used for acquiring image data and point cloud data, and sending the image data and point cloud data to an image processor after preprocessing; and performing corresponding operations on the vehicle according to a processing result of the image processor on the image data and point cloud data.

[0009] The image processor searches for abnormal behaviors affecting the running of the vehicle in the smallest search range by using a target detection algorithm after acquiring batch image data, extracts feature points from the point cloud data, and performs a calculation task in the running process of the vehicle.

[0010] Further, the information acquisition unit includes a vehicle-mounted camera and a laser radar; the vehicle-mounted camera and the laser radar are both connected to the central processor.

[0011] The vehicle-mounted camera is used for acquiring image data in the running process of the vehicle.

[0012] The laser radar is used for acquiring point cloud data in the running process of the vehicle.

[0013] Further, the preprocessing process of the central processor on the image data and the point cloud data includes:

[0014] The central processor generates batch image data by matrix splicing the image data, and saves the batch image data in the memory; and generates batch point cloud data by matrix splicing the point cloud data, and saves the batch point cloud data in the memory.

[0015] Further, the image processor includes a first image processor and a second image processor; and the first image processor and the second image processor are both in communication connection with the central processor.

[0016] The first image processor is used for acquiring batch image data, and searching for abnormal behaviors affecting the running of the vehicle in the smallest search range by using a target detection algorithm.

[0017] The second image processor is used for extracting feature points from the point cloud data, sending the point cloud data after feature point extraction to the central processor, and performing a calculation task in the running process of the vehicle according to a processing result of the central processor.

[0018] Further, the central processor performs target matching calculation processing on the point cloud data after extracting feature points, and identifies a target type; and sends the target type to the image processor after identifying the target type.

[0019] The image processor calculates a speed to be converted, an acceleration to be converted, and an angle of turning according to the identified target type and the searched abnormal behaviors affecting the running of the vehicle.

[0020] Further, the process of searching for abnormal behaviors affecting vehicle operation in the smallest search range by the first image processor using a target detection algorithm comprises:

[0021] defining a sub-region in the region of image data as a search region where an abnormal object is likely to be found;

[0022] defining a search factor to make the search region grow in four directions with equal probability, and the search region is bounded by the region of image data, and the abnormal behaviors affecting vehicle operation are searched.

[0023] Further, the system further comprises a fault-tolerant module;

[0024] The fault-tolerant module is connected to the central processor and the image processor, and the central processor and the image processor both send heartbeat signals to the fault-tolerant module at regular intervals, and if the fault-tolerant module does not receive the heartbeat signals within a timeout period, it is determined that the system has failed.

[0025] The application also proposes a method for improving the performance of a vehicle-mounted computing platform, which is realized based on a system for improving the performance of a vehicle-mounted computing platform, and comprises the following steps:

[0026] obtaining image data and point cloud data during vehicle driving;

[0027] preprocessing the image data and point cloud data by the central processor, and then sending them to the image processor;

[0028] obtaining batch image data by the image processing module, searching for abnormal behaviors affecting vehicle operation in the smallest search range by using a target detection algorithm, extracting feature points from the point cloud data, and performing calculation tasks during vehicle driving;

[0029] performing target matching on the point cloud data after extracting feature points by the central processor, and performing corresponding operations on the vehicle in combination with the abnormal behaviors affecting vehicle operation.

[0030] Further, the process of searching for abnormal behaviors affecting vehicle operation in the smallest search range by the first image processor using a target detection algorithm comprises:

[0031] defining a sub-region in the region of image data as a search region where an abnormal object is likely to be found;

[0032] defining a search factor to make the search region grow in four directions with equal probability, and the search region is bounded by the region of image data, and the abnormal behaviors affecting vehicle operation are searched.

[0033] Further, the heartbeat signals of the image processor and the heartbeat signals of the central processor are obtained at regular intervals to determine whether the vehicle-mounted computing platform has failed.

[0034] The effects provided in the summary are only the effects of the embodiments, not all the full effects of the invention, and one of the technical solutions has the following advantages or beneficial effects:

[0035] The present application provides a system and method for improving the performance of a vehicle-mounted computing platform, which includes an information acquisition unit, a central processor and an image processor. The information acquisition unit is used to acquire image data and point cloud data during vehicle driving. The central processor is used to obtain the image data and point cloud data, and to send the preprocessed image data and point cloud data to the image processor. The image processor obtains batch image data, and uses a target detection algorithm to search for abnormal behaviors affecting vehicle operation in the smallest search range. Feature points are extracted from the point cloud data, and calculation tasks during vehicle driving are performed. Based on a system for improving the performance of a vehicle-mounted computing platform, a method for improving the performance of a vehicle-mounted computing platform is also proposed. The present application adds a fault-tolerant mechanism to the vehicle-mounted computing platform to improve the fault tolerance of the vehicle-mounted computing platform, and uses a central processor and an image processor to perform collaborative computation to maximize the computing power of the vehicle-mounted computing platform.

[0036] The present application uses a target detection algorithm to search for abnormal behaviors affecting vehicle operation in the smallest search range, selects the position of the most likely obstacle to define the search factor, and can achieve good performance and efficiency in practical applications. In addition to reducing the identification time, reducing the identification area can also reduce the possibility of false matching. BRIEF DESCRIPTION OF DRAWINGS

[0037] As Figure 1 Fig. 1 is a connection diagram of a system for improving the performance of a vehicle-mounted computing platform according to an embodiment of the present application;

[0038] As Figure 2 Fig. 2 is a search area diagram of a target recognition algorithm according to an embodiment of the present application;

[0039] As Figure 3 Fig. 3 is a flowchart of a method for improving the performance of a vehicle-mounted computing platform according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0041] Example 1

[0042] Embodiment 1 of this invention proposes a system for improving the performance of an in-vehicle computing platform. It improves the fault tolerance of the in-vehicle computing platform by adding a fault-tolerant mechanism and maximizes the computing power of the in-vehicle computing platform by using a method of collaborative operation of a central processing unit and a graphics processing unit.

[0043] The CPU is the central processing unit of a computer, playing a crucial role in controlling the computer's operation. The GPU, on the other hand, is a secondary processor that primarily handles graphics-related tasks and presents the data more effectively on the display. In essence, the GPU is the heart of the graphics card, similar to the role of the CPU in a computer. Compared to the CPU, the GPU has several advantages: (1) abundant computing resources; (2) smaller control components. Therefore, the GPU is only suitable for tasks with few branches, large data volumes, and low latency. The CPU, however, excels at logical control and handles tasks with many branches.

[0044] The CPU has high processing speed but few computing units, and the GPU has low processing speed but many computing units; in the prior art, the GPU is used for calculation, and the CPU is idle; however, resources are limited, and when a large amount of logic and data need to be calculated, the GPU is not powerful enough, and the Hermes vehicle-mounted computing platform has strict requirements on the amount of calculation, and it is hoped that the system computing power can be maximized in limited physical resources. The application atomizes the calculation operation, and the program is divided into two parts: one part is executed in the GPU, and the other part is executed in the CPU. Specifically, when there is a calculation task, a CPU main thread and multiple CPU slave threads are first started from the CPU end, the main thread is responsible for processing input parameters and completing corresponding initialization operations, and the main thread is responsible for scheduling GPU devices and CPU devices, the GPU device is responsible for calculating the high-density graphics rendering task, and the CPU device is responsible for performing the transaction calculation task with strong logic; the system resources are maximized by parallel calculation; considering that the data transmission bandwidth between the GPU and the CPU is limited, the data transmission is reduced during configuration, the overall communication load is reduced, and the optimization efficiency is improved.

[0045] As Figure 1 It is a system connection schematic diagram for improving the performance of a vehicle-mounted computing platform according to Embodiment 1 of the application, and the system comprises an information acquisition unit, a central processing unit and an image processor.

[0046] The information acquisition unit is used for acquiring image data and point cloud data in the driving process of a vehicle.

[0047] The central processing unit is used for acquiring the image data and the point cloud data, sending the image data and the point cloud data to the image processor after preprocessing, and performing corresponding operations on the vehicle according to the processing result of the image processor on the image data and the point cloud data.

[0048] After the image processor acquires batch image data, an abnormal behavior affecting the operation of the vehicle is searched in the smallest search range by using a target detection algorithm, feature points of the point cloud data are extracted, and a calculation task in the driving process of the vehicle is performed.

[0049] The information acquisition unit comprises a vehicle-mounted camera and a laser radar; the vehicle-mounted camera and the laser radar are both connected to the central processing unit; the vehicle-mounted camera is used for acquiring image data in the driving process of the vehicle; and the laser radar is used for acquiring point cloud data in the driving process of the vehicle.

[0050] Lidar is a product of the combination of laser technology, electronic technology and signal processing technology. High-frequency laser can obtain a large amount of (106-107 order of magnitude) position point information (referred to as point cloud) in one second. The distance information of the detected object can be calculated by using the time of the return of the projection to the detected object, and a three-dimensional model can be drawn according to the information to draw a 3D environment map, and the precision can reach centimeter level, thereby improving the measurement accuracy. Different surface feature information will also be generated according to the different reflectivity of the object surface. The point cloud can be regarded as an image, and powerful deep learning technology can be used to detect the objects thereon.

[0051] The process of pre-processing the image data and the point cloud data by the central processor includes: the central processor performs matrix splicing on the image data to generate batch image data, and saves the batch image data in the memory; and performs matrix splicing on the point cloud data to generate batch point cloud data, and saves the batch point cloud data in the memory.

[0052] The image processor includes a first image processor and a second image processor; and the first image processor and the second image processor are both in communication connection with the central processor.

[0053] The first image processor is used to acquire batch image data, and uses a target detection algorithm to search for abnormal behaviors affecting vehicle operation in the smallest search range.

[0054] The second image processor is used to extract feature points from the point cloud data, and sends the point cloud data after feature point extraction to the central processor, and executes a calculation task in a vehicle driving process according to a processing result of the central processor.

[0055] The GPU is designed as a coprocessor, and is usually connected with the CPU through a PCI-e bus. The relatively low bandwidth and high delay of the PCI-e bus are usually bottleneck problems. In order to efficiently and fully utilize the limited computing resources on the mobile processor to achieve high performance, it is necessary to divide the tasks between the CPU and the GPU, explore the parallelism of the algorithm, and optimize the memory access.

[0056] Task division: OpenCL is a programming framework designed for heterogeneous computing across various platforms. In OpenCL, a host processor (usually a CPU) manages the OpenCL context and can assign parallel tasks to multiple computing devices (such as GPUs, CPUs). Parallel jobs can be divided into multiple workgroups, each consisting of many work items, which are the basic processing units of parallel execution kernels. OpenCL defines a hierarchical memory model, which includes a large off-chip global memory with long latency of hundreds of clock cycles, and a small but fast on-chip local memory. With the support of the OpenCL parallel programming model, we can perform hybrid computing on CPU+GPU; the workflow after task division is as follows:

[0057] The CPU reads the image data and point cloud data of the image to be processed into memory;

[0058] The CPU performs matrix splicing on the image data to generate batch image data, which is saved in memory, and performs matrix splicing on the point cloud data to generate batch point cloud data, which is saved in memory;

[0059] The GPU reads the batch image data into the video memory and the batch point cloud data into the video memory;

[0060] The GPU performs target detection based on the OpenCL-implemented target detection algorithm; and performs feature point extraction on the batch point cloud data; or performs a computing task

[0061] The GPU reads the detected target into memory; and reads the point cloud data after feature point extraction into memory;

[0062] The CPU performs matching calculation and processing on the target data to identify abnormal behaviors that affect vehicle operation; and performs target recognition based on the point cloud data after feature point extraction, so that the vehicle performs corresponding operations such as turning, accelerating, decelerating, or avoiding, etc. For example: detecting a pedestrian, and calculating whether it will hit him; detecting a vehicle in front, and calculating its speed, whether to slow down or suddenly change lanes, etc.

[0063] In this application, the GPU includes a first GPU and a second GPU. The first GPU and the second GPU also perform collaborative operation. The first GPU is used to perform tasks such as target detection and feature point extraction, and the second GPU performs computing tasks in the vehicle driving process in addition to target detection and feature point extraction.

[0064] Computer vision as a kind of through the computer help or replace the human eye perception of image or multidimensional data technology, widely used in the field of unmanned driving. And in the field of unmanned driving, the target tracking and detection based on vision plays an important role. Through the data obtained by the vehicle-mounted camera in the computer for target detection, image classification, feature detection and matching data calculation.

[0065] The process of the target tracking algorithm based on vision of the present application is to define a sub-region that may find abnormal objects in the region of image data as a search region;

[0066] Define a search factor, so that the search region grows in four directions with equal probability, and the region of image data as the boundary, search for abnormal behavior affecting vehicle operation.

[0067] As Figure 2 The search region schematic diagram of the target recognition algorithm of embodiment 1 of the present application; wherein M and N are the height and width of the image obtained by the vehicle-mounted camera,

[0068] Target detection algorithm: mainly to detect whether there is an abnormality affecting vehicle operation from the photo taken by the camera; in order to reduce the time and find the abnormality faster, define the width and height of the region most likely to find the object as w and h, so that w*h is the search region of the object; but this region is only the region that may search for the object, so this algorithm also designs a search factor α, when no abnormality is found, the search region grows in four directions with equal opportunity, thereby increasing the possibility of identifying the target.

[0069] The units of M, N, w, h and α here are pixels.

[0070] This algorithm is actually an algorithm for optimizing target recognition, searching for abnormalities in the smallest search range, and the actual search area is smaller than the height* width of the picture.

[0071] The size of the search region is:

[0072] Among them, B left is the left boundary of the search region; B right is the right boundary of the search region; B top is the upper boundary of the search region; B bottom is the lower boundary of the search region, and α is the search factor.

[0073] The size of the search area is: SA=(2α+1)w·(2α+1)h-wh=((2α+1)2-1)wh.

[0074] When we reduce the recognition area (reduce alpha), the complexity of recognizing the target is significantly reduced. After reducing the search factor, the best matching patch is limited to a small area around the object area. Selecting a suitable search factor alpha is crucial for the actual application to achieve good performance and efficiency, and selecting the position where the obstacle is most likely to appear can define the search factor alpha to make the actual application achieve good performance and efficiency. In addition to reducing the recognition time, reducing the recognition area can also reduce the possibility of false matching.

[0075] In order to deal with the downtime caused by faults or other reasons in the calculation process, and improve the stability of the system, a fault diagnosis and recovery architecture is added to improve the stability of the system; the CPU and the GPU send heartbeat information to the fault diagnosis module in time, and if the heartbeat information is not received within a certain time, it is judged that a fault has occurred, and when the fault is determined, the vehicle-mounted computing platform is recovered, and if the recovery fails, the restart processing is performed.

[0076] The system for improving the performance of the vehicle-mounted computing platform provided in Embodiment 1 of the present application improves the fault tolerance of the vehicle-mounted computing platform by adding a fault tolerance mechanism in the vehicle-mounted computing platform, and maximizes the computing power of the vehicle-mounted computing platform by using the central processor and the image processor for collaborative operation.

[0077] The system for improving the performance of the vehicle-mounted computing platform provided in Embodiment 1 of the present application uses a target detection algorithm to search for abnormal behaviors that affect vehicle operation in the smallest search range, selects the position where the obstacle is most likely to appear as the search factor, and can make the actual application achieve good performance and efficiency. In addition to reducing the recognition time, reducing the recognition area can also reduce the possibility of false matching.

[0078] Embodiment 2

[0079] Based on the system for improving the performance of the vehicle-mounted computing platform provided in Embodiment 1 of the present application, Embodiment 2 of the present application further provides a method for improving the performance of the vehicle-mounted computing platform. As Figure 3 The flowchart of the method for improving the performance of the vehicle-mounted computing platform in Embodiment 1 of the present application.

[0080] In step S300, image data and point cloud data during vehicle driving are acquired;

[0081] The information acquisition unit includes a vehicle-mounted camera and a laser radar; the vehicle-mounted camera and the laser radar are both connected to the central processor; the vehicle-mounted camera is used to acquire image data during vehicle driving; and the laser radar is used to acquire point cloud data during vehicle driving.

[0082] In step S310, the image data and the point cloud data of the to-be-processed image are read into the memory by the central processor CPU;

[0083] In step S320, the image data is matrix spliced by the central processing unit CPU to generate batch image data, the batch image data is saved in the memory, and the point cloud data is matrix spliced to generate batch point cloud data, and the batch point cloud data is saved in the memory;

[0084] In step S330, the batch image data is read into the video memory and the batch point cloud data is read into the video memory by the image processor GPU;

[0085] In step S340, the target detection algorithm based on the implementation of OpenCL is used by the image processor GPU to perform target detection; and the batch point cloud data is subjected to feature point extraction; or a computing task is performed.

[0086] Wherein M and N are the height and width of the image obtained by the vehicle-mounted camera,

[0087] Target detection algorithm: mainly to detect whether there is an abnormality affecting the operation of the vehicle from the photo taken by the camera; in order to reduce the time and find the abnormality faster, define the width and height of the region where the object is most likely to be found as w and h, so that w*h is the search area of the object; but this area is only the area where the object may be searched, so the algorithm also designs a search factor a, when no abnormality is found, the search area increases in four directions with equal opportunity, thereby increasing the possibility of identifying the target.

[0088] The units of M, N, w, h and a here are pixels

[0089] This algorithm is actually an algorithm for optimizing the identification of targets, searching for abnormalities in the smallest search range, and the actual search area is smaller than the height*width of the picture.

[0090] The size of the search area is:

[0091] Wherein, B left is the left boundary of the search area; B right is the right boundary of the search area; B top is the upper boundary of the search area; B bottom is the lower boundary of the search area, and a is the search factor.

[0092] The size of the search area is: SA=(2a+1)w·(2a+1)h-wh=((2a+1)2-1)wh.

[0093] When we reduce the recognition area (reduce alpha), the complexity of recognizing the target is significantly reduced. After reducing the search factor, the best matching patch is limited in a small area around the object area. Selecting a suitable search factor alpha is crucial for the actual application to achieve good performance and efficiency. In addition to reducing the recognition time, reducing the recognition area can also reduce the possibility of false matching.

[0094] The GPU also performs extraction of point-wise features from the original point cloud data, and performs computing tasks after target detection and feature point extraction, etc.

[0095] In the present application, the GPU includes a first GPU and a second GPU. The first GPU and the second GPU also perform collaborative operation. The first GPU is used to perform tasks such as target detection and feature point extraction, and the second GPU performs computing tasks during vehicle driving in addition to target detection and feature point extraction.

[0096] In step S350, the detected target is read into the memory by the image processor GPU; and the point cloud data after feature point extraction is read into the memory;

[0097] In step S360, the target data is processed by the central processing unit CPU for matching calculation, and the abnormal behavior affecting the vehicle operation is recognized; and the target is recognized according to the point cloud data after feature point extraction, so that the vehicle makes corresponding operations such as turning, accelerating, decelerating or avoiding, etc. For example: detecting a pedestrian, it is necessary to calculate whether it will hit him; detecting a vehicle in front, it is necessary to calculate whether his speed will be reduced or suddenly change lanes, etc.

[0098] In the process of running, the heartbeat signals of the image processor and the central processing unit are obtained at regular intervals, and it is judged whether the vehicle-mounted computing platform fails.

[0099] The method for improving the performance of the vehicle-mounted computing platform proposed in embodiment 2 of the present application adds a fault-tolerant mechanism to the vehicle-mounted computing platform to improve the fault tolerance of the vehicle-mounted computing platform, and adopts a method of collaborative operation of the central processing unit and the image processor to maximize the computing power of the vehicle-mounted computing platform.

[0100] The method for improving the performance of the vehicle-mounted computing platform proposed in embodiment 2 of the present application uses a target detection algorithm to search for abnormal behaviors affecting the vehicle operation in the smallest search range, selects the position where the obstacle is most likely to appear as the search factor, so that the actual application program can achieve good performance and efficiency. In addition to reducing the recognition time, reducing the recognition area can also reduce the possibility of false matching.

[0101] The method for improving performance of a vehicle-mounted computing platform proposed in embodiment 2 of the present application is implemented based on the system for improving performance of a vehicle-mounted computing platform proposed in embodiment 1. If the description of the method in embodiment 2 is not detailed enough, reference should be made to the process in embodiment 1.

[0102] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Furthermore, the article "a" or "an" is intended to include the plural, unless the context clearly indicates otherwise. The above-described embodiments of the present application provide technical solutions identical to the corresponding technical solutions in the prior art. However, the above-described technical solutions in the embodiments of the present application are not described in detail to avoid excessive redundancy.

[0103] The above describes the specific embodiments of the present application in conjunction with the accompanying drawings, but is not intended to limit the scope of protection of the present application. Based on the above description, those skilled in the art can make other different forms of modifications or changes. Here, it is not necessary or possible to exhaust all the embodiments. Various modifications or changes made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. A system for improving performance of an in-vehicle computing platform, the system comprising: The system comprises an information acquisition unit, a central processor and an image processor. The information acquisition unit is configured to acquire image data and point cloud data during vehicle driving. The central processor is configured to acquire the image data and the point cloud data, pre-process the image data and the point cloud data, and send the pre-processed image data and point cloud data to the image processor. The image processor is configured to acquire batch image data, search for abnormal behaviors affecting vehicle driving in a minimum search range by using a target detection algorithm, extract feature points from the point cloud data, and perform a calculation task during vehicle driving according to a processing result of the image processor. The image processor comprises a first image processor and a second image processor, and the first image processor and the second image processor are both in communication connection with the central processor. The first image processor is configured to acquire batch image data, and search for abnormal behaviors affecting vehicle driving in a minimum search range by using a target detection algorithm. The second image processor is configured to extract feature points from the point cloud data, send the point cloud data after feature point extraction to the central processor, and perform a calculation task during vehicle driving according to a processing result of the central processor. where the size of the search region is: where B left is the left boundary of the search area; B right is the right boundary of the search area; B top is the top boundary of the search area; B bottom is the bottom boundary of the search area, and a is a search factor. The process that the first image processor searches for abnormal behaviors affecting vehicle driving in a minimum search range by using a target detection algorithm comprises the following steps.

2. The system for improving performance of an in-vehicle computing platform of claim 1, wherein, A sub-region in which an abnormal object is likely to be found is defined as a search region in a region of the image data. A search factor is defined to make the search region grow in four directions with equal probability, and the image data region is taken as a boundary to search for abnormal behaviors affecting vehicle driving. The size of the search region is SA=(2α+1)w·(2α+1)h-wh=((2α+1)2-1)wh.

3. The system for improving performance of an in-vehicle computing platform of claim 1, wherein, The information acquisition unit comprises a vehicle-mounted camera and a laser radar, and the vehicle-mounted camera and the laser radar are both connected to the central processor. The vehicle-mounted camera is configured to acquire image data during vehicle driving.

4. The system for improving performance of an in-vehicle computing platform of claim 1, wherein, The laser radar is configured to acquire point cloud data during vehicle driving. The process that the central processor pre-processes the image data and the point cloud data comprises the following steps.

5. The system for improving performance of an in-vehicle computing platform of any one of claims 1 to 4, wherein, The central processor performs matrix splicing on the image data to generate batch image data, saves the batch image data in a memory, performs matrix splicing on the point cloud data to generate batch point cloud data, and saves the batch point cloud data in the memory. The central processor performs target matching calculation and processing on the point cloud data after feature point extraction to identify a target type, and sends the target type to the image processor.

6. A method for improving the performance of an in-vehicle computing platform, implemented based on the system for improving the performance of an in-vehicle computing platform according to any one of claims 1 to 5, characterized in that, The image processor calculates a speed, an acceleration and a turning angle to be converted according to the identified target type and the searched abnormal behaviors affecting vehicle driving. The system further comprises a fault-tolerant module. The fault-tolerant module is connected to the central processor and the image processor, and the central processor and the image processor both send heartbeat signals to the fault-tolerant module at regular time intervals. The system further comprises the following steps. Acquire image data and point cloud data during vehicle driving. The image data and the point cloud data are preprocessed by a central processor and then sent to an image processor; Batch image data are acquired by an image processing module, an abnormal behavior affecting vehicle operation is searched in a smallest search range by using a target detection algorithm, feature points of the point cloud data are extracted, and a calculation task in a vehicle driving process is executed; The central processor performs target matching on the point cloud data after the feature points are extracted, and in combination with the abnormal behavior affecting the vehicle operation, performs corresponding operations on the vehicle.

7. The method of improving performance of an in-vehicle computing platform of claim 6, wherein, The heartbeat signals of the image processor and the central processor are acquired in a timely manner, and it is judged whether the vehicle-mounted computing platform has a fault.

Images