TWI710997B - Auto insurance automatic compensation method and system - Google Patents

Auto insurance automatic compensation method and system Download PDF

Info

Publication number
TWI710997B
TWI710997B TW108136377A TW108136377A TWI710997B TW I710997 B TWI710997 B TW I710997B TW 108136377 A TW108136377 A TW 108136377A TW 108136377 A TW108136377 A TW 108136377A TW I710997 B TWI710997 B TW I710997B
Authority
TW
Taiwan
Prior art keywords
user
insurance
credit
image
damage
Prior art date
Application number
TW108136377A
Other languages
Chinese (zh)
Other versions
TW202025060A (en
Inventor
周凡
Original Assignee
開曼群島商創新先進技術有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 開曼群島商創新先進技術有限公司 filed Critical 開曼群島商創新先進技術有限公司
Publication of TW202025060A publication Critical patent/TW202025060A/en
Application granted granted Critical
Publication of TWI710997B publication Critical patent/TWI710997B/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/20—Administration of product repair or maintenance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Finance (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • Multimedia (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Technology Law (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

本發明揭示了一種用於車險自動賠付的方法。該方法包括引導用戶拍攝車輛圖像並獲取該車輛圖像；對所獲取的車輛圖像進行圖像識別以標識車輛損傷；基於該車輛損傷來產生維修清單；在該用戶接受該維修清單的情況下審查用戶信用；以及在該用戶信用足夠高的情況下自動賠付該用戶。The invention discloses a method for automatic compensation of auto insurance. The method includes guiding a user to take a vehicle image and obtain the vehicle image; performing image recognition on the obtained vehicle image to identify the vehicle damage; generating a repair list based on the vehicle damage; and when the user accepts the repair list Under review the user’s credit; and automatically compensate the user if the user’s credit is sufficiently high.

Description

車險自動賠付方法和系統Auto insurance automatic compensation method and system

本發明涉及金融科技，尤其涉及一種用於車險自動賠付的方法和系統。The present invention relates to financial technology, and in particular to a method and system for automatic payment of auto insurance.

據統計，當今中國每年約有4500萬件私家車保險索賠案，其中60%左右為“純外觀損傷”，即2700萬件。目前，每單處理成本約為150元，這為各個保險公司帶來了承重的查勘成本。 而且，車主發生事故時，車主通常會給保險公司打電話報案，由保險公司安排查勘人員到現場進行查勘定損，或由車主對損傷處進行拍照後將照片發給保險公司，在經過現場或遠端查勘後，確定需要維修更換的項目，再由車主去維修廠進行修理，拿到發票或維修清單之後，將這些材料交給保險公司，保險公司對所有相關單據進行審核後，將維修費用賠付給車主。整個過程比較冗長，車主需要關注的事項較多，拿到賠付款的速度也很慢。 另外，為給車主提供便利，部分保險公司提供了代收單據的服務，但也需要保險公司花費相當的人工成本，也無法從根本上縮短車主拿到賠付款的時間。此外，國內保險行業也會採用小額案件快速賠付的方案，對於金額低於一定閾值的案件，提前將費用賠付給車主，之後再審核各類材料，但這又會帶來一定的欺詐風險，對部分案件可能造成不合理賠付。 現有的賠付流程如果希望能夠做到過程嚴謹控制，減少不合理賠付比例，就會使得賠付流程較長，手續繁瑣，車主體驗不好；如果希望減少車主麻煩，簡化流程，快速賠付，又會提高不合理賠付比例。 當前，金融科技(英語：Financial technology，也稱為FinTech)的快速發展使得解決車險行業的高查勘成本、低效賠付流程和理賠欺詐問題成為可能。金融科技是人工智慧(AI)、深度學習等技術在金融產業取得突破性發展的大背景下應運而生的產物。如何在車險場景這片熱土上，最大程度挖掘 AI 技術演算法以及資料的價值和潛力，成為保險行業內一眾玩家努力開墾的新方向。 因此，保險公司或相關企業透過運用科技手段(尤其是AI技術和演算法)來使得理賠流程變得更有效率並且降低查勘成本和不合理賠付率是合乎需要的。According to statistics, there are about 45 million private car insurance claims in China each year, of which about 60% are "pure cosmetic damage", or 27 million. At present, the processing cost per order is about 150 yuan, which has brought load-bearing survey costs to various insurance companies. Moreover, when a car owner has an accident, the car owner usually calls the insurance company to report the case, and the insurance company arranges survey personnel to investigate the damage on the spot, or the car owner takes a picture of the damage and sends the photo to the insurance company. After the remote survey, the items that need to be repaired and replaced are determined, and the owner will go to the repair shop for repairs. After obtaining the invoice or repair list, the materials will be handed over to the insurance company. The insurance company will review all relevant documents and pay the repair costs Pay to the owner. The whole process is relatively lengthy, car owners need to pay attention to more matters, and the speed of getting compensation is very slow. In addition, in order to provide convenience to car owners, some insurance companies provide the service of collecting documents, but it also requires insurance companies to spend considerable labor costs, and it cannot fundamentally shorten the time for car owners to receive compensation. In addition, the domestic insurance industry will also adopt a small-amount case quick payment plan. For cases where the amount is less than a certain threshold, the cost will be paid to the car owner in advance, and various materials will be reviewed later, but this will bring a certain risk of fraud. Some cases may cause unreasonable compensation. If the existing compensation process hopes to achieve rigorous process control and reduce the proportion of unreasonable compensation, the compensation process will be longer, the procedures are cumbersome, and the owner's experience is not good; if you want to reduce the trouble of the owner, simplify the process, and pay quickly, it will increase Unreasonable compensation ratio. At present, the rapid development of financial technology (English: Financial technology, also known as FinTech) makes it possible to solve the problems of high investigation costs, inefficient claims procedures and claims fraud in the auto insurance industry. Financial technology is the product of artificial intelligence (AI), deep learning and other technologies that have emerged as the times require breakthroughs in the financial industry. How to maximize the value and potential of AI technology algorithms and data in the hot land of the auto insurance scene has become a new direction for players in the insurance industry to develop. Therefore, it is desirable for insurance companies or related companies to use technological means (especially AI technology and algorithms) to make the claims process more efficient and reduce survey costs and unreasonable loss ratios.

提供本發明內容來以簡化形式介紹將在以下具體實施方式部分中進一步描述的一些概念。本發明內容並不旨在標識出所要求保護的主題的關鍵特徵或必要特徵，也不旨在用於幫助確定所要求保護的主題的範圍。 對保險公司而言，為了根據事故現場做出損傷原因的初步判斷並確認損傷的車零件及損傷程度，往往要雇傭大量的人力長期駐派多地進行重複的拍照作為定損記錄的依據。本發明提供的技術方案則依託深度學習圖像演算法逐步替代定損環節中的重複性人工作業流程，將大大降低車險定損環節中的人力以及時間成本。 具體而言，針對保險理賠領域中的高查勘成本、理賠流程繁瑣和理賠欺詐這些問題，本發明提供了一種車險自動賠付方案以解決上述缺陷。具體地，本發明希望透過圖像識別、人工智慧、深度學習、網際網路支付技術及用戶信用體系，提供一種針對純外觀案件全自動賠付的方法，既能實現賠付款實時到帳，避免車主和查勘員參與的流程或手續過多，又能控制賠付的精准程度以防止欺詐行為。 在本發明的一個實施例中，提供了一種用於車險自動賠付的方法，該方法包括： 獲取車輛圖像； 對所獲取的車輛圖像進行圖像識別以標識車輛損傷； 基於車輛損傷來產生維修清單； 審查用戶信用；以及 自動賠付用戶。 在本發明的一個實施例中，用戶信用是在用戶接受維修清單的情況下審查的，並且該方法還包括在用戶不接受該維修清單的情況下重新獲取車輛圖像。在本發明的一個實施例中，自動賠付是在用戶信用足夠高的情況下執行的並且該方法還包括在用戶信用不夠高的情況下將用戶的本次理賠流程轉為人工處理。 在本發明的一個實施例中，提供了一種用於車險自動賠付的系統，該系統包括： 用於獲取車輛圖像的裝置； 用於對所獲取的車輛圖像進行圖像識別以標識車輛損傷的裝置； 用於基於車輛損傷來產生維修清單的裝置； 用於審查用戶信用的裝置；以及 用於自動賠付用戶的裝置。 在本發明的一個實施例中，用戶信用是在用戶接受維修清單的情況下審查的，並且該系統還包括用於在用戶不接受該維修清單的情況下重新獲取車輛圖像的裝置。在本發明的一個實施例中，自動賠付是在用戶信用足夠高的情況下執行的，並且該系統還包括用於在用戶信用不夠高的情況下將用戶的本次理賠流程轉為人工處理的裝置。 在本發明的一個實施例中，提供了一種儲存用於車險自動賠付的指令的電腦可讀取儲存媒體，這些指令包括： 用於獲取車輛圖像的指令； 用於對所獲取的車輛圖像進行圖像識別以標識車輛損傷的指令； 用於基於車輛損傷來產生維修清單的指令； 用於審查用戶信用的指令；以及 用於自動賠付用戶的指令。 在本發明的一個實施例中，用戶信用是在用戶接受維修清單的情況下審查的，並且該電腦可讀取儲存媒體還包括用於在用戶不接受該維修清單的情況下重新獲取車輛圖像的指令。在本發明的一個實施例中，自動賠付是在用戶信用足夠高的情況下執行的，並且該電腦可讀取儲存媒體還包括用於在用戶信用不夠高的情況下將用戶的本次理賠流程轉為人工處理的指令。 本發明的各方面一般包括如基本上在本文參照圖式所描述並且透過圖式所闡示的方法、裝置、系統、電腦程式產品。 在結合圖式研讀了下文對本發明的具體示例性實施例的描述之後，本發明的其他方面、特徵和實施例對於本領域普通技術人員將是明顯的。儘管本發明的特徵在以下可能是針對某些實施例和圖式來討論的，但本發明的全部實施例可包括本文所討論的有利特徵中的一個或多個。換言之，儘管可能討論了一個或多個實施例具有某些有利特徵，但也可以根據本文討論的本發明的各種實施例使用此類特徵中的一個或多個特徵。以類似方式，儘管示例性實施例在下文可能是作為設備、系統或方法實施例進行討論的，但是應當領會，此類示例性實施例可以在各種設備、系統、和方法中實現。The content of the present invention is provided to introduce some concepts that will be further described in the following detailed description in a simplified form. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to help determine the scope of the claimed subject matter. For insurance companies, in order to make a preliminary judgment of the cause of the damage and confirm the damaged car parts and the extent of the damage based on the scene of the accident, they often need to hire a large number of manpower to be stationed in multiple locations for repeated photographs as the basis for damage assessment records. The technical scheme provided by the present invention relies on the deep learning image algorithm to gradually replace the repetitive manual operation process in the loss assessment link, which will greatly reduce the manpower and time cost in the loss assessment link of car insurance. Specifically, in view of the problems of high investigation costs, cumbersome claims process and claims fraud in the field of insurance claims, the present invention provides an auto insurance automatic claim payment solution to solve the above-mentioned defects. Specifically, the present invention hopes to provide a fully automatic compensation method for pure appearance cases through image recognition, artificial intelligence, deep learning, Internet payment technology, and user credit system, which can achieve real-time payment of compensation and avoid car owners. There are too many processes or procedures involved with investigators, and the accuracy of compensation can be controlled to prevent fraud. In an embodiment of the present invention, there is provided a method for automatic payment of auto insurance, and the method includes: Acquire vehicle images; Perform image recognition on the acquired vehicle images to identify vehicle damage; Generate maintenance list based on vehicle damage; Review user credit; and Pay users automatically. In an embodiment of the present invention, the user credit is reviewed when the user accepts the maintenance list, and the method further includes reacquiring the vehicle image when the user does not accept the maintenance list. In an embodiment of the present invention, the automatic claim payment is performed when the user's credit is sufficiently high, and the method further includes converting the user's current claims settlement process to manual processing when the user's credit is not high enough. In an embodiment of the present invention, there is provided a system for automatic payment of auto insurance, which includes: A device for acquiring vehicle images; A device for performing image recognition on the acquired vehicle image to identify vehicle damage; Device for generating maintenance list based on vehicle damage; Devices used to review user credit; and A device used to automatically pay users. In an embodiment of the present invention, the user credit is reviewed when the user accepts the maintenance list, and the system further includes a device for reacquiring the vehicle image when the user does not accept the maintenance list. In an embodiment of the present invention, the automatic payment is executed when the user's credit is sufficiently high, and the system also includes a method for converting the user's current claims process to manual processing when the user's credit is not high enough Device. In an embodiment of the present invention, there is provided a computer-readable storage medium storing instructions for automatic payment of auto insurance. These instructions include: Instructions for obtaining vehicle images; Instructions for image recognition of the acquired vehicle image to identify vehicle damage; Instructions for generating maintenance lists based on vehicle damage; Instructions for reviewing user credit; and Instructions used to automatically pay users. In an embodiment of the present invention, the user credit is reviewed when the user accepts the maintenance list, and the computer-readable storage medium further includes a method for reacquiring the vehicle image when the user does not accept the maintenance list. Instructions. In an embodiment of the present invention, the automatic compensation is executed when the user's credit is sufficiently high, and the computer-readable storage medium also includes the current compensation process for the user when the user's credit is not high enough Converted to manual processing instructions. Aspects of the present invention generally include methods, devices, systems, and computer program products as basically described herein with reference to the drawings and illustrated through the drawings. After studying the following description of specific exemplary embodiments of the present invention in conjunction with the drawings, other aspects, features and embodiments of the present invention will be apparent to those of ordinary skill in the art. Although the features of the present invention may be discussed below with respect to certain embodiments and drawings, all embodiments of the present invention may include one or more of the advantageous features discussed herein. In other words, although one or more embodiments may be discussed as having certain advantageous features, one or more of such features may also be used in accordance with the various embodiments of the invention discussed herein. In a similar manner, although exemplary embodiments may be discussed below as device, system, or method embodiments, it should be appreciated that such exemplary embodiments may be implemented in various devices, systems, and methods.

以下將參考形成本發明一部分並示出各具體示例性實施例的圖式更詳盡地描述各個實施例。然而，各實施例可以以許多不同的形式來實現，並且不應將其解釋為限制此處所闡述的各實施例；相反地，提供這些實施例以使得本發明變得透徹和完整，並且將這些實施例的範圍完全傳達給本領域普通技術人員。各實施例可按照方法、系統或設備來實施。因此，這些實施例可採用硬體實現形式、全軟體實現形式或者結合軟體和硬體方面的實現形式。因此，以下具體實施方式並非是局限性的。 圖1A、1B、2及相關聯的描述提供了其中可實施本發明的各實施例的各種操作環境的討論。然而，關於圖1A、1B、2所示出和討論的設備和系統是用於示例和說明的目的，而非對可被用於實施本文所述的本發明的各實施例的大量計算設備配置的限制。 圖1A和1B示出可用來實施本發明的各實施例的合適的行動計算環境，例如行動電話、智慧型電話、輸入板個人電腦、膝上型電腦等。參考圖1A，示出了用於實現各實施例的示例行動計算設備100。在一基本配置中，行動計算設備100是具有輸入元件和輸出元件兩者的手持式電腦。輸入元件可包括允許用戶將資訊輸入到行動計算設備100中的觸控螢幕顯示器105和輸入按鈕110。行動計算設備100還可結合允許進一步的用戶輸入的可選的側面輸入元件115。可選的側面輸入元件115可以是旋轉開關、按鈕、或任何其他類型的手動輸入元件。在替代實施例中，行動計算設備100可結合更多或更少的輸入元件。例如，在某些實施例中，顯示器105可以不是觸控螢幕。在又一替代實施例中，行動計算設備是可攜式電話系統，如具有顯示器105和輸入按鈕110的蜂巢式電話。行動計算設備100還可包括可選的小鍵盤135。可選的小鍵盤135可以是實體小鍵盤或者在觸控螢幕顯示器上產生的“軟”小鍵盤。 行動計算設備100結合輸出元件，如可顯示圖形用戶介面(GUI)的顯示器105。其他輸出元件包括揚聲器125和LED 120。另外，行動計算設備100可包含振動模組(未示出)，該振動模組使得行動計算設備100振動以將事件通知給用戶。在又一實施例中，行動計算設備100可結合耳機插孔(未示出)，用於提供另一手段來提供輸出信號。 儘管此處組合行動計算設備100來描述，但在替代實施例中，本發明還可組合任何數量的電腦系統來被使用，如在台式環境中、膝上型或筆記本電腦系統、多處理器系統、基於微處理器或可程式化消費電子產品、網路PC、小型電腦、大型電腦等。本發明的實施例也可在分布式計算環境中實踐，其中任務由分布式計算環境中透過通信網路鏈接的遠端處理設備來執行；程式可位於本機和遠端記憶體儲存設備中。總而言之，具有多個環境傳感器、向用戶提供通知的多個輸出元件和多個通知事件類型的任何電腦系統可結合本發明的實施例。 圖1B是示出在一個實施例中使用的諸如圖1A中所示的計算設備之類的行動計算設備的組件的方塊圖。即，行動計算設備100可結合系統102以實現某些實施例。例如，系統102可被用於實現可運行與台式或筆記本電腦的應用程式類似的一個或多個應用程式的“智慧型電話”，這些應用程式例如演示文稿應用程式、瀏覽器、電子郵件、日程安排、即時訊息收發、以及媒體播放器應用程式。在某些實施例中，系統102被集成為計算設備，諸如集成的個人數位助理(PDA)和無線電話。 一個或多個應用程式166可被加載到記憶體162中並在作業系統164上或與作業系統164相關聯地運行。應用程式的示例包括電話撥號程式、電子郵件程式、PIM(個人資訊管理)程式、文字處理程式、電子表格程式、網際網路瀏覽器程式、訊息通信程式等等。系統102還包括記憶體168內的非揮發性儲存162。非揮發性儲存168可被用於儲存在系統102斷電時不會丟失的持久資訊。應用程式166可使用資訊並將資訊儲存在非揮發性儲存168中，如電子郵件應用程式使用的電子郵件或其他訊息等。同步應用程式(未示出)也可駐留在系統102上並被程式化為與駐留在主機電腦上的對應同步應用程式進行交互，以保持儲存在非揮發性儲存168中的資訊與儲存在主機電腦上的對應資訊相同步。如應被理解的，其他應用程式可被加載到記憶體162中且在設備100上運行，包括車險理賠應用程式26。 系統102具有可被實現為一個或多個電池的電源170。電源170還可包括外部功率源，如補充電池或對電池重新充電的AC適配器或加電對接托架。 系統102還可包括執行發射和接收無線電頻率通信的功能的無線電172。無線電172透過通信營運商或服務供應商方便了系統102與“外部世界”之間的無線連接。來往無線電172的傳輸是在作業系統164的控制下進行的。換言之，無線電172接收的通信可透過作業系統164傳播到應用程式166，反之亦然。 無線電172允許系統102例如透過網路與其他計算設備通信。無線電172是通信媒體的一個示例。通信媒體由諸如載波或其他傳輸機制等已調變資料信號中的電腦可讀取指令、資料結構、程式模組或其他資料來體現，並包括任何資訊傳遞媒體。術語“已調變資料信號”是指使得以在信號中編碼資訊的方式來設置或改變其一個或多個特性的信號。作為示例而非限制，通信媒體包括諸如有線網路或直接線連接之類的有線媒體，以及諸如聲學、RF、紅外及其他無線媒體之類的無線媒體。如此處所使用的術語電腦可讀取媒體包括儲存媒體和通信媒體兩者。 系統102的該實施例是以兩種類型的通知輸出設備來示出的：可被用於提供視覺通知的LED 120，以及可被用於揚聲器125提供音頻通知的音頻介面174。這些設備可直接耦合到電源170，使得當被激活時，即使為了節省電池功率而可能關閉處理器160和其他組件，它們也在一段由通知機制指示的持續時間保持通電。LED 120可被程式化為無限地保持通電，直到用戶採取行動指示該設備的通電狀態。音頻介面174用於向用戶提供聽覺信號並從用戶接收聽覺信號。例如，除被耦合到揚聲器125以外，音頻介面174還可被耦合到話筒以接收聽覺輸入，諸如便於電話對話。根據各本發明的各實施例，話筒也可充當音頻傳感器來便於對通知的控制，如下文將描述的。系統102可進一步包括允許板載相機130的操作來記錄靜止圖像、視頻流等的視頻介面176。 行動計算設備實現系統102可具有附加特徵或功能。例如，該設備還可包括附加資料儲存設備(可移動的/或不可移動的)，諸如磁盤、光盤或磁帶。此類附加儲存在圖1B中由儲存168示出。電腦儲存媒體可包括以用於儲存諸如電腦可讀取指令、資料結構、程式模組、或其他資料等資訊的任何方法或技術實現的揮發性和非揮發性、可移動和不可移動媒體。 設備100產生或捕捉的且經系統102儲存的資料/資訊可如上所述本地儲存在設備100上，或資料可被儲存在可由設備透過無線電172或透過設備100和與設備100相關聯的分開的計算設備之間的有線連接存取的任何數量的儲存媒體上，該分開的計算設備如例如網際網路之類的分布式計算網路中的伺服器電腦。如應理解的，此類資料/資訊可經設備100、經無線電172或經分布式計算網路來被存取。類似地，這些資料/資訊可根據已知的資料/資訊傳送和儲存手段來容易地在計算設備之間傳送以儲存和使用，這些手段包括電子郵件和協作資料/資訊共享系統。 圖2示出了其中可實現本發明的各實施例的聯網環境。用戶持有平板計算設備204或者行動計算設備206，平板計算設備204或者行動計算設備206中的每一者都包括本文描述的車險理賠應用程式客戶端202。用戶可操作車險理賠應用程式客戶端202並透過包括但不限於網際網路的網路208來與車險理賠應用程式伺服器端212通信。 車險理賠應用程式伺服器端212在伺服器210中實現並且包括用於為用戶提供車險理賠服務，具體而言用於進行圖像分析、車損決策、理賠金額計算等操作的定損模型214。車險理賠應用程式伺服器端212還包括用於儲存各種操作資料或模型資料的儲存216。 車險理賠應用程式伺服器端212還與N個保險公司218對接以便從各個保險公司接收各種保險資料(包括但不限於零配件資料、車型資料、賠付率資料、出險記錄資料、維修資料等等)以及向保險公司反饋理賠決策。 圖3示出了根據本發明的一個實施例的車險理賠應用程式300的方塊圖。該車險理賠應用程式300用於快速地為用戶的車損事故提供定損方案，該功能透過車險理賠應用程式客戶端302和車險理賠應用程式伺服器端310的組合來實現。車險理賠應用程式300的示例包括但不限於螞蟻金服公司開發的“定損寶”。 車險理賠應用程式客戶端302作為一個應用程式(APP)被安裝在用戶擁有的行動計算設備或平板計算設備上，並且該車險理賠應用程式客戶端302也可以是用戶的行動計算設備或平板計算設備上已安裝的現有應用程式的插件。 車險理賠應用程式客戶端302包括UI呈現模組304、圖像獲取模組306、以及通信模組308。UI呈現模組304透過用戶的行動計算設備或平板計算設備的顯示器來向用戶呈現圖形用戶介面以引導用戶拍攝車損圖像並上傳車損圖像、獲得定損資訊以及確認理賠資訊。 圖像獲取模組306用於獲取車損圖像資訊，具體而言該圖像獲取模組306可透過UI呈現模組來引導用戶透過安裝有車險理賠應用程式客戶端302的行動計算設備或平板計算設備來拍攝車損圖像，包括包含車牌資訊的全車圖像、車損部位的遠景圖像、車損部位的近景圖像、以及車損細節圖像。如果用戶的車輛具有不止一處損壞，則可引導用戶針對每一處損壞分別拍攝車損部位的遠景圖像、車損部位的近景圖像、以及車損細節圖像，以完整地記錄全部車損資訊。 另外，圖像獲取模組306還能夠確定用戶拍攝的圖像是否滿足自動損傷識別的要求，即確定用戶拍攝的圖像是否可供車險理賠應用程式自動識別車損資訊。如果確定用戶拍攝的圖像不滿足自動損傷識別的要求，則引導用戶重新拍攝相應圖像以達到自動損傷識別的要求。 在本發明的一個實施例中，圖像獲取模組306還能夠引導用戶拍攝車損視頻，以使得能夠獲取更豐富的車損資訊以便於自動定損。 通信模組308使得車險理賠應用程式客戶端302能夠與車險理賠應用程式伺服器端310通信，以便將透過圖像獲取模組306獲取的圖像傳送到車險理賠應用程式伺服器端310以供進行進一步處理。 車險理賠應用程式伺服器端310駐留在車險自動理賠服務提供商的伺服器處，在本發明的一個實施例中，該車險自動理賠服務由螞蟻金服公司提供。但在其他實施例中，該車險自動理賠服務也可由其他公司提供。 車險理賠應用程式伺服器端310包括通信模組312、圖像識別模組314、定損模組316、信用審查模組318、賠付模組320、以及定損模型322。 通信模組312用於從車險理賠應用程式客戶端302接收車損圖像資料並將這些資料傳遞至圖像識別模組314。 圖像識別模組314用於對所接收到的車損圖像進行圖像識別。具體而言，首先對接收到的車損圖像進行過濾，即基於是否滿足自動損傷識別要求來過濾掉不符合要求的圖像，並且透過通信模組312來通知車險理賠應用程式客戶端302以引導用戶重新拍攝相應圖像。圖像識別模組314在接收到所有必需的車損圖像後對這些車損圖像進行雜訊去除、零件識別、損傷檢測、原因判斷、以及程度判定。這些功能將在下文中更詳細地描述，並且這些功能都基於車險理賠應用程式伺服器端310所包含的定損模型322。 然而，在使用定損模型322來進行圖像識別之前，該定損模型322的訓練和學習是一項重大挑戰。近年來，深度學習以及電腦視覺技術獲得了長足發展，在部分簡單的任務上(比如ImageNet分類)甚至達到了比人更高的精確度，但是面對車險定損這樣一個複雜的現實場景，演算法的攻堅道路仍是困難重重，在車險定損領域依然鮮見有效的技術方案落地。在現實場景中，演算法需要應對多種複雜光照條件、污漬、水滴、車型構造等多種干擾因素，並從海量資料中學習到對定損有效的關鍵資訊。 為此，在本發明的一個實施例中，首先對來自各個保險公司的千萬級雜亂無章的車險定損歷史圖片進行結構化規整、資料整理、清洗以及必要的標注，這個龐大圖像資料庫的照片數量以及標簽的複雜程度對比現有的ImageNet都要高出一個數量級。隨後，將經整理、清洗和標注的車險定損歷史圖片提供給定損模型322以供在基於ASIC、FPGA、GPU等晶片技術的軟硬一體化異構機器學習平台上進行學習和訓練。該定損模型322利用已有的沉澱的大量歷史定損資料, 針對不同的車型、顏色和光照條件進行模型迭代學習，最終能夠輸出較為精准的零件識別結果以及針對多種程度的刮擦、變形、零件的開裂和脫落等損傷的定損結論。 具體而言，定損模型322使用深度神經網路檢測車輛受損部位及其在圖像中的區域。在本發明的一個示例性且非限制性實施例中，可以基於卷積神經網路(Convolutional Neural Network，CNN)和區域建議網路(Region Proposal Network，RPN)，結合池化層、全連接層等來預先構建該定損模型322，然後透過來自各保險公司的大量經整理、清洗和標注的車險定損歷史圖片來訓練該定損模型322以產生深度神經網路。在其它實施例中，還可結合全連接層(Fully-Connected Layer，FC)、池化層、資料歸一化層等。在另一實施例中，如果需要對受損部位進行分類，還可以在定損模型322中加入概率輸出層(Softmax)等。 卷積神經網路一般指以卷積層為主要結構並結合其他如激活層等組成的神經網路，主要用於圖像識別。本實施例中所述的深度神經網路可以包括卷積層和其他重要的層(如池化層，資料歸一化層，激活層等)，並結合區域建議網路(RPN)共同組建產生。卷積神經網路通常是將圖像處理中的二維離散卷積運算和人工神經網路相結合。這種卷積運算可以用於自動提取特徵。區域建議網路(RPN)可以將一個圖像(任意大小)提取的特徵作為輸入(可以使用卷積神經網路提取的二維特徵)，輸出矩形目標建議方塊的集合，每個方塊有一個對象的得分。在另一實施例中，可以將所使用的卷積神經網路(CNN)稱為卷積層(CNN)、將區域建議網路(RPN)稱為區域建議層(RPN)。在本發明的其它實施例中，定損模型322還可以結合基於該卷積神經網路改進後的或區域建議網路改進後的變種網路模型，經過樣本資料訓練後構建所產生的深度卷積神經網路，即定損模型322。 上述實施例中使用的模型和演算法可以選擇同類模型或者演算法。具體地，例如在定損模型322中，可以使用基於卷積神經網路和區域建議網路的多種模型和變種，如Faster R-CNN、Y0L0、Mask-FCN等。其中的卷積神經網路(CNN)可以用任意CNN模型，如ResNet、Inception、VGG等及其變種。通常神經網路中的卷積網路部分可以使用在物體識別取得較好效果的成熟網路結構，如Inception、ResNet等網路。在ResNet網路中，輸入為一張圖片，輸出為多個含有損傷部位的圖片區域以及對應的損傷分類(損傷分類用於確定損傷類型)和置信度(置信度為表示損傷類型真實性程度的參量)。Faster R-CNN、Y0L0、Mask-FCN 等都是屬於本實施例可以使用的包含卷積層的深度神經網路。本文描述的定損模型322能夠結合區域建議層和CNN層檢測出受損部位、損傷類型和程度、以及受損部位在該零件圖像中所處的位置區域。 回到圖3，隨後圖像識別模組314將識別結果傳遞至定損模組316，該定損模組316進而將定損結果或定損明細資訊傳送到相應的保險公司，即用戶購買其車險的保險公司，以便由該保險公司基於車輛識別碼(VIN碼)(基於車牌號獲取)去各自的資料庫中查找受損零件的OE碼，然後根據OE碼就能找到當地的零件維修和更換的價格，並結合來自圖像識別模組314的定損明細資訊來產生相應的維修清單。隨後保險公司將相應的維修清單傳送回定損模組316。每個保險公司的報價可能不一樣，最後出的價格也可能不一樣。 定損模組316然後將維修清單透過通信模組312傳遞至車險理賠應用程式客戶端302，車險理賠應用程式客戶端302透過UI呈現模組304向用戶呈現該維修清單。 圖4示出了由UI呈現模組304呈現給用戶的圖形用戶介面400的示例。可以理解，本發明的各實施例不限於該示例性用戶介面400。如圖所示，UI呈現模組304向用戶顯示本地事故定損總金額以及維修清單，如果用戶認可該維修清單，則該用戶可激活“接受”按鈕404以繼續至下一步驟，即信用審查步驟。如果用戶不認可該維修清單(例如有損傷未識別，或覺得應該進行花費更高的維修)，則可選擇重新拍攝，即該用戶可激活“重新拍攝”按鈕406以返回到車損圖像拍攝步驟以用於重新拍攝車損圖像並重新定損。或者，用戶可放棄自動理賠，轉入人工處理，如同傳統車險報案和理賠流程。 另外，用戶還可激活定損總金額右側的“來年保費預測”按鈕402，以基於本地出險、本年度出險次數、理賠金額、個人信用等因素來透過車險理賠應用程式預測下一年度的保費。此處的個人信用可以由第三方信用服務提供商來給出，可以理解第三方信用服務提供商包括但不限於螞蟻金服公司，且個人信用包括但不限於由螞蟻金服公司開發的車險分，這將在下文中更詳細地描述。如本領域技術人員可以理解的，圖4所示的用戶介面400僅僅是出於闡示的目的而是示例性且非限制性的。 在用戶認可所呈現的維修清單，即在用戶激活“接受”按鈕404的情況下，信用審查模組318審查用戶的個人信用以確定是進行自動即時賠付還是轉人工處理。在本發明的一個實施例中，車險理賠應用程式伺服器端310採用由螞蟻金服公司開發的車險分來確定用戶的個人信用分，但應理解，還可以採用其他信用體系或等級，諸如螞蟻金服公司開發的芝麻信用等。 傳統上，要預測客戶下一年的保險金額或賠付率，車險行業一般的做法是基於客戶上一年度的出險次數、區域、車型、車價、使用性質等因素來計算，這些因素都屬於車因素。而車險分還能夠基於人(比如性別、年齡、職業、身份)、行為(諸如違章次數、是否經常上高速、信用歷史、消費習慣、駕駛習慣)和使用環境(比如道路類型、道路擁擠狀況)三大方面的因素來更精准地預測客戶下一年的保險金額或賠付率，從而極大地豐富了車險中“與人相關的”資料維度。車險分透過對車主進行精准畫像和風險分析得出例如300-700不等的分數，分數越高，風險越低。低風險的車主一般都有良好的駕駛習慣、信用行為。 保險公司在獲得用戶授權的情況下，可以查詢用戶的車險標準分，或是結合自身資料對標簽進行加工建模，得到自己的車險專用分，從而依據車險分進行更為公平的車險定價。具體地，透過快速進行海量資料標簽挖掘以及用於提升預測性能的機器學習演算法來以提高的模型迭代速度融合保險公司累積的車險理賠資料以及信用服務提供商累積的用戶畫像資料(都經過脫敏處理)以產生用戶的車險分。 回到對信用審查模組318的描述，當確定用戶的信用足夠高時，信用審查模組318可通知賠付模組320進行自動即時賠付，即在沒有人工審核的情況下自動將定損總金額匯入用戶指定的銀行帳戶或線上帳戶。作為示例而非限制，信用足夠高可以指信用或其某一表現形式達到或超過特定閾值或等級，諸如車險分超過600分。 在本發明的一個實施例中，如果完成自動賠付，但事後在審核材料階段發現車主有欺詐行為，則保險公司將降低車主的信用，以防止車主持續進行欺詐。而且，保險公司也可與第三方中立平台合作，將此次索賠相關資訊經過符合法律要求的脫敏處理後，傳送到第三方平台以進行跨保險公司的信用記錄徵集，並在多個保險公司之間共享，以防止車主在多家保險公司進行欺詐索賠。此處描述的第三方平台包括但不限於中國保信。 如果確定用戶信用分不夠高，則信用審查模組318通知賠付模組320將用戶本次理賠流程轉為人工處理，如同傳統的車險理賠流程。 圖5示出了根據本發明的另一實施例的車險理賠應用程式的另一示例。在本發明的該另一實施例中，不同於拍攝車損圖像並將其傳送到伺服器端做技術處理再返回到客戶端，還可以引導用戶圍繞車輛拍攝視頻以提供更豐富的車輛圖像資訊，並且得益於近年來智慧型手機的處理器性能和圖像處理能力的大幅提高，可以將車險理賠應用程式伺服器端310的功能整合到車險理賠應用程式客戶端302以使得客戶端302能夠自動對車出現的損傷做出即時分析。具體而言，可以在用戶拍攝視頻時透過AR(增強實境)技術實時疊加損傷判定和維修資訊，從而直接且即時地告訴用戶損傷的程度，向用戶自動推薦維修清單，從而能大幅提高拍攝引導、反饋的時效性，甚至做到實時定損。如圖5中的圖示502所示，在用戶拍攝視頻時車險理賠應用程式能夠實時地引導用戶拍攝更高質量的視頻圖像，諸如引導用戶更靠近損傷部位以便做出更精確的車損判定。另外，如圖5中的圖示504所示，在用戶拍攝視頻時做出實時定損判斷之際還能夠結合損傷原因判斷做出諸如“疑似非本案損傷”等更高階的智能判定。而且，由於不將用戶車輛圖像資訊上傳至伺服器端而是利用用戶持有的行動計算設備或平板計算設備的本地CPU和GPU的能力，因此能一定程度上解決用戶隱私、資訊安全等方面的問題。 圖6示出了根據本發明的一個實施例的圖像識別模組314的詳細功能方塊圖。在本發明的一個實施例中，圖像識別模組314首先透過圖像過濾組件602對接收到的車損圖像進行過濾，即基於是否滿足自動損傷識別要求來過濾掉不符合要求的圖像，並且透過通信模組312來通知車險理賠應用程式客戶端302以引導用戶重新拍攝相應圖像。例如，當用戶拍攝了車輛全景圖像並上傳後，圖像識別模組314中的圖像過濾組件602可確定該全景圖像是否包括車牌、是否取景過遠，等等。作為另一示例，當用戶拍攝了車損細節圖像並上傳後，圖像過濾組件602可確定該車損細節圖像是否完整地包括車損細節、細節是否清晰可見，等等。如果上傳的圖像不符合自動損傷識別要求，則圖像過濾組件602過濾掉該圖像並透過UI呈現模組304引導用戶重新拍攝相應圖像。 當已經接收到自動損傷識別所必需的所有圖像後，圖像識別模組314中的雜訊去除組件604使用定損模型322來標識並排除諸如光反射、陰影、倒影、污漬、水滴、車型構造等多種雜訊干擾因素。定損模型322已經透過所沉澱的大量歷史定損資料(尤其是帶雜訊的圖像資料)進行訓練和學習，並由此能夠準確地標識出各種干擾因素並將其排除。如圖7中的圖示702所示，示出了基於定損模型322的反光雜訊去除，在經過雜訊去除組件604的處理後，車輛左前葉子板上的反光被去除。 然後，零件識別組件606基於車損部位的遠景圖像透過定損模型322來識別車輛的各個零件，包括受損零件。具體地，零件識別指的是從一輛車的圖像中檢測出各個零件的類別和位置，如識別圖片中哪裡是前機蓋、左前大燈、保險桿、格柵等等。在本發明的一個實施例中，基於定損模型322結合Faster R-CNN(更快的區域性卷積神經網路)等網路技術來完成零件識別，但應理解在其他實施例中還可以結合其他技術來完成零件標識。 在標識出車輛的各個零件後，損傷檢測組件608透過經大量樣本訓練的定損模型322來檢測損傷部位和類別。損傷部位指的是受損的具體零件，損傷類別包括例如刮擦、變形、開裂、脫落等。如果包含損傷部位的圖像的拍攝角度不佳，對角度進行矯正。具體地，由於用戶一般不會經過專業訓練，自然不可能保證每一張上傳的照片都拍得端正明晰，因此有時需要對圖像進行矯正，從而能夠更好地進行損傷檢測、原因判斷和損傷程度判定。矯正技術包括但不限於基於投影的方法、基於Hough變換、基於線性擬合，以及傅裡葉變換。圖7中的圖示704示出了包含損傷部位的圖像的矯正示例。 隨後原因判斷組件610透過經大量樣本訓練的定損模型322來判斷損傷原因，諸如單車刮擦、銳器損傷、雙車刮擦、雙車碰撞等。另外，原因判斷組件610還能識別出疑似非本案損傷，例如基於損傷顏色顯著不同或者損傷部位與碰撞部位不符，等等。 最後，程度判定組件612透過經大量樣本訓練的定損模型322來判定受損程度，諸如輕微刮擦(不露底漆)、嚴重刮擦(露底漆)、輕微變形、嚴重變形、輕微開裂、嚴重開裂、報廢，等等，以便產生相應的維修清單。此處描述的定損模型322已經在上文中詳細描述，因此在此不再贅述。本發明的實施例不限於上述受損程度。 圖8示出了根據本發明的一個實施例的用於車險自動賠付的方法800的流程圖。在各實施例中，圖8所示的步驟可透過硬體(例如，處理器、引擎、記憶體、電路)、軟體(例如，作業系統、應用程式、驅動器、機器/處理器可執行指令)或其組合來執行。如本領域普通技術人員將理解的，各實施例可以包括比示出的更多或更少的步驟。 在802，引導用戶拍攝車輛圖像並獲取車輛圖像。可透過車險理賠應用程式客戶端302內的UI呈現模組來引導用戶透過安裝有車險理賠應用程式客戶端302的行動計算設備或平板計算設備來拍攝車損圖像，包括包含車牌資訊的全車圖像、車損部位的遠景圖像、車損部位的近景圖像、以及車損細節圖像。另外，還能夠確定用戶拍攝的圖像是否滿足自動損傷識別的要求，並且在確定用戶拍攝的圖像不滿足自動損傷識別要求的情況下引導用戶重新拍攝相應圖像以達到自動損傷識別要求。在本發明的另一實施例中，還能夠引導用戶拍攝車損視頻，以使得能夠獲取更豐富的車損資訊以便於自動實時定損。 在方塊804，對所獲取的車輛圖像進行圖像識別以標識車輛損傷。圖像識別包括圖像過濾、雜訊去除、零件識別、損傷檢測、原因判斷、以及程度判定，這些功能步驟將在圖9中更詳細地描述。在本發明的一個實施例中，這些圖像識別功能透過基於深度神經網路的經大量樣本訓練的定損模型來完成。 在方塊806，基於車輛損傷來產生維修清單。車輛損傷包括損傷部位、損傷類別和損傷程度。針對零件輕度刮擦可採用噴漆維修，針對零件變形可採用板金維修，針對零件開裂、脫落等嚴重損傷可直接更換零件。在將定損結果或定損明細資訊傳送到相應的保險公司後，該保險公司基於車輛識別碼(VIN碼)(基於車牌號獲取)去各自的資料庫中查找受損零件的OE碼，然後根據OE碼就能找到當地的零件維修和更換的價格，並產生相應的維修清單。 在方塊808，詢問用戶是否接受所產生的維修清單。如果用戶接受，則流程繼續至方塊810，否則流程返回至方塊802以重新引導用戶拍攝車輛圖像。 在方塊810，審查用戶的個人信用。個人信用可結合保險公司提供的理賠資料以及由第三方公司提供的個人信用資料。在本發明的一個實施例中，用戶的個人信用可採取由螞蟻金服公司開發的車險分的形式，但應理解可採取其它信用評估形式。 在方塊812，確定用戶信用是否足夠高以做出是進行自動即時賠付還是轉人工處理的判斷。如果確定用戶信用足夠高，則流程繼續至方塊814，自動賠付用戶。否則，流程繼續至方塊816，將本次理賠專由按照傳統理賠流程的人工處理。 圖9示出了根據本發明的一個實施例的用於圖像識別的方法900的流程圖。在各實施例中，圖9所示的步驟可透過硬體(例如，處理器、引擎、記憶體、電路)、軟體(例如，作業系統、應用程式、驅動器、機器/處理器可執行指令)或其組合來執行。如本領域普通技術人員將理解的，各實施例可以包括比示出的更多或更少的步驟。 在902，對接收到的車損圖像進行過濾。換言之，基於是否滿足自動損傷識別要求來過濾掉不符合要求的圖像，並且在確定圖像不符合要求的情況下過濾掉該圖像並引導用戶重新拍攝相應圖像。 在904，對經過濾的車損圖像進行雜訊去除。具體地，諸如光反射、陰影、倒影、污漬、水滴、車型構造等多種雜訊干擾因素被標識並排除。 在906，標識車輛的各個零件，包括受損零件。該標識基於車損部位的遠景圖像，並且從一輛車的圖像中檢測出各個零件的類別和位置，如識別圖片中哪裡是前機蓋、左前大燈、保險桿、格柵等等。 在908，檢測車損圖像中的損傷部位和類別以確定具體受損零件以及受損類別。在本發明的一個實施例中，如果包含損傷部位的圖像的拍攝角度不佳，對角度進行矯正以使得能夠更好地進行損傷檢測、原因判斷和損傷程度判定。 在910，確定損傷原因。損傷原因包括單車刮擦、銳器損傷、雙車刮擦、雙車碰撞等。另外，還能識別出疑似非本案損傷，例如基於損傷顏色顯著不同或者損傷部位與碰撞部位不符，等等。 在912，判定受損程度以便產生相應的維修清單。受損程度包括輕微刮擦、嚴重刮擦、輕微變形、嚴重變形、輕微開裂、嚴重開裂、報廢，等等。 以上參考根據本發明的實施例的方法、系統和電腦程式產品的方塊圖和/或操作說明描述了本發明的實施例。方塊中所注明的各功能/動作可以按不同於任何流程圖所示的次序出現。例如，取決於所涉及的功能/動作，連續示出的兩個方塊實際上可以基本上同時執行，或者這些方塊有時可以按相反的次序來執行。 以上說明、示例和資料提供了對本發明的組成部分的製造和使用的全面描述。因為可以在不背離本發明的精神和範圍的情況下做出本發明的許多實施例，所以本發明落在所附申請專利範圍的範圍內。Hereinafter, each embodiment will be described in more detail with reference to the drawings forming a part of the present invention and showing each specific exemplary embodiment. However, the embodiments can be implemented in many different forms, and should not be construed as limiting the embodiments set forth herein; on the contrary, these embodiments are provided to make the present invention thorough and complete, and to combine these The scope of the embodiments is fully conveyed to those of ordinary skill in the art. Each embodiment can be implemented according to a method, system or device. Therefore, these embodiments may adopt a hardware implementation form, an all-software implementation form, or a combination of software and hardware implementation forms. Therefore, the following specific embodiments are not limited. Figures 1A, 1B, 2 and the associated description provide a discussion of various operating environments in which various embodiments of the invention can be implemented. However, the devices and systems shown and discussed with respect to FIGS. 1A, 1B, and 2 are for purposes of example and description, and are not configured for a large number of computing devices that can be used to implement the various embodiments of the present invention described herein. limits. 1A and 1B illustrate suitable mobile computing environments, such as mobile phones, smart phones, tablet PCs, laptop computers, etc., that can be used to implement various embodiments of the present invention. Referring to Figure 1A, an example mobile computing device 100 for implementing various embodiments is shown. In a basic configuration, the mobile computing device 100 is a handheld computer with both input elements and output elements. The input element may include a touch screen display 105 and an input button 110 that allow the user to input information into the mobile computing device 100. The mobile computing device 100 may also incorporate an optional side input element 115 that allows further user input. The optional side input element 115 may be a rotary switch, a button, or any other type of manual input element. In alternative embodiments, the mobile computing device 100 may incorporate more or fewer input elements. For example, in some embodiments, the display 105 may not be a touch screen. In yet another alternative embodiment, the mobile computing device is a portable telephone system, such as a cellular phone with a display 105 and an input button 110. The mobile computing device 100 may also include an optional keypad 135. The optional keypad 135 may be a physical keypad or a "soft" keypad generated on a touch screen display. The mobile computing device 100 incorporates output elements, such as a display 105 that can display a graphical user interface (GUI). Other output elements include speakers 125 and LED 120. In addition, the mobile computing device 100 may include a vibration module (not shown) that causes the mobile computing device 100 to vibrate to notify the user of the event. In yet another embodiment, the mobile computing device 100 may be combined with a headphone jack (not shown) to provide another means to provide an output signal. Although described here in combination with the mobile computing device 100, in alternative embodiments, the present invention can also be used in combination with any number of computer systems, such as in a desktop environment, laptop or notebook computer system, and multi-processor system. , Microprocessor-based or programmable consumer electronics products, network PCs, small computers, large computers, etc. The embodiments of the present invention can also be practiced in a distributed computing environment, in which tasks are executed by remote processing equipment linked through a communication network in the distributed computing environment; programs can be located in local and remote memory storage devices. In summary, any computer system with multiple environmental sensors, multiple output elements to provide notifications to users, and multiple notification event types can incorporate embodiments of the present invention. Figure 1B is a block diagram illustrating components of a mobile computing device such as the computing device shown in Figure 1A used in one embodiment. That is, the mobile computing device 100 may be combined with the system 102 to implement certain embodiments. For example, the system 102 can be used to implement a "smart phone" that can run one or more applications similar to those on a desktop or laptop computer, such as presentation applications, browsers, emails, and calendars. Scheduling, instant messaging, and media player applications. In some embodiments, the system 102 is integrated as a computing device, such as an integrated personal digital assistant (PDA) and wireless phone. One or more application programs 166 may be loaded into the memory 162 and run on or in association with the operating system 164. Examples of application programs include telephone dialing programs, email programs, PIM (Personal Information Management) programs, word processing programs, spreadsheet programs, Internet browser programs, messaging programs, etc. The system 102 also includes non-volatile storage 162 in the memory 168. The non-volatile storage 168 can be used to store persistent information that will not be lost when the system 102 is powered off. The application 166 can use the information and store the information in a non-volatile storage 168, such as email or other messages used by an email application. A synchronization application (not shown) may also reside on the system 102 and be programmed to interact with a corresponding synchronization application that resides on the host computer to maintain the information stored in the non-volatile storage 168 and the host computer The corresponding information on the computer is synchronized. As should be understood, other applications can be loaded into the memory 162 and run on the device 100, including the auto insurance claims application 26. The system 102 has a power source 170 that can be implemented as one or more batteries. The power source 170 may also include an external power source, such as an AC adapter or a power docking cradle to supplement or recharge the battery. The system 102 may also include a radio 172 that performs the function of transmitting and receiving radio frequency communications. The radio 172 facilitates the wireless connection between the system 102 and the "outside world" through the communication operator or service provider. The transmission to and from the radio 172 is carried out under the control of the operating system 164. In other words, the communication received by the radio 172 can be propagated to the application 166 via the operating system 164, and vice versa. The radio 172 allows the system 102 to communicate with other computing devices via a network, for example. The radio 172 is an example of a communication medium. The communication medium is embodied by computer-readable instructions, data structures, program modules or other data in a modulated data signal such as carrier waves or other transmission mechanisms, and includes any information transmission medium. The term "modulated data signal" refers to a signal that allows one or more of its characteristics to be set or changed in a manner that encodes information in the signal. By way of example and not limitation, communication media includes wired media such as wired networks or direct wire connections, and wireless media such as acoustic, RF, infrared, and other wireless media. The term computer readable media as used herein includes both storage media and communication media. This embodiment of the system 102 is shown with two types of notification output devices: an LED 120 that can be used to provide visual notifications, and an audio interface 174 that can be used to provide audio notifications by a speaker 125. These devices may be directly coupled to the power source 170 so that when activated, even if the processor 160 and other components may be turned off in order to save battery power, they remain powered on for a duration indicated by the notification mechanism. The LED 120 can be programmed to remain powered on indefinitely until the user takes action to indicate the power-on status of the device. The audio interface 174 is used to provide auditory signals to and receive auditory signals from the user. For example, in addition to being coupled to the speaker 125, the audio interface 174 may also be coupled to a microphone to receive auditory input, such as to facilitate telephone conversations. According to various embodiments of the present invention, the microphone can also act as an audio sensor to facilitate the control of notifications, as will be described below. The system 102 may further include a video interface 176 that allows the operation of the onboard camera 130 to record still images, video streams, and the like. The mobile computing device implementation system 102 may have additional features or functions. For example, the device may also include additional data storage devices (removable/or non-removable) such as magnetic disks, optical disks, or tapes. Such additional storage is shown by storage 168 in FIG. 1B. Computer storage media may include volatile and non-volatile, removable and non-removable media implemented by any method or technology used to store information such as computer readable instructions, data structures, program modules, or other data. The data/information generated or captured by the device 100 and stored by the system 102 can be stored locally on the device 100 as described above, or the data can be stored in a separate device that can be used by the device via radio 172 or via the device 100 and associated with the device 100 On any number of storage media accessed by wired connections between computing devices, such separate computing devices are server computers in a distributed computing network such as the Internet. As should be understood, such data/information can be accessed via the device 100, via the radio 172, or via a distributed computing network. Similarly, these data/information can be easily transmitted between computing devices for storage and use based on known data/information transmission and storage means, including e-mail and collaborative data/information sharing systems. Figure 2 shows a networking environment in which various embodiments of the present invention can be implemented. The user holds a tablet computing device 204 or a mobile computing device 206, and each of the tablet computing device 204 or the mobile computing device 206 includes the auto insurance claims application client 202 described herein. The user can operate the auto insurance claims application client 202 and communicate with the auto insurance claims application server 212 through a network 208 including but not limited to the Internet. The auto insurance claims application server 212 is implemented in the server 210 and includes a loss determination model 214 for providing users with auto insurance claims services, specifically for image analysis, car loss decision-making, and claim amount calculations. The auto insurance claims application server 212 also includes a storage 216 for storing various operating data or model data. The auto insurance claims application server 212 also connects with N insurance companies 218 to receive various insurance data from each insurance company (including but not limited to parts data, model data, loss rate data, insurance record data, maintenance data, etc.) And feedback the claims decision to the insurance company. FIG. 3 shows a block diagram of a car insurance claims application program 300 according to an embodiment of the present invention. The auto insurance claims application 300 is used to quickly provide a loss determination solution for the user's car accident. This function is implemented through the combination of the auto insurance claims application client 302 and the auto insurance claims application server 310. Examples of the auto insurance claims application program 300 include, but are not limited to, "Dingxibao" developed by Ant Financial. The auto insurance claim application client 302 is installed as an application (APP) on a mobile computing device or tablet computing device owned by the user, and the auto insurance claim application client 302 can also be the user's mobile computing device or tablet computing device Plug-ins for existing applications that have been installed. The auto insurance claims application client 302 includes a UI presentation module 304, an image acquisition module 306, and a communication module 308. The UI presentation module 304 presents a graphical user interface to the user through the display of the user's mobile computing device or tablet computing device to guide the user to take car damage images and upload the car damage images, obtain damage assessment information, and confirm claims information. The image acquisition module 306 is used to acquire car damage image information. Specifically, the image acquisition module 306 can guide the user through the mobile computing device or tablet installed with the car insurance claims application client 302 through the UI presentation module The computing device takes the car damage image, including the whole car image containing the license plate information, the long-range image of the damaged part, the close-up image of the damaged part, and the detailed image of the car damage. If the user’s vehicle has more than one damage, the user can be guided to take the long-range image of the damaged part, the close-up image of the damaged part, and the detailed image of the damage for each damage, so as to fully record all the cars. Loss of information. In addition, the image acquisition module 306 can also determine whether the image taken by the user meets the requirements of automatic damage recognition, that is, whether the image taken by the user can be used for the auto insurance claims application to automatically recognize the car damage information. If it is determined that the image taken by the user does not meet the requirements of automatic damage recognition, the user is guided to retake corresponding images to meet the requirements of automatic damage recognition. In an embodiment of the present invention, the image acquisition module 306 can also guide the user to shoot a car damage video, so that more abundant car damage information can be obtained for automatic damage determination. The communication module 308 enables the auto insurance claims application client 302 to communicate with the auto insurance claims application server 310, so as to send the images acquired through the image acquisition module 306 to the auto insurance claims application server 310 for processing Further processing. The auto insurance claims application program server 310 resides at the server of the auto insurance automatic claims service provider. In an embodiment of the present invention, the auto insurance automatic claims service is provided by Ant Financial. However, in other embodiments, the auto insurance automatic claims service may also be provided by other companies. The auto insurance claims application server 310 includes a communication module 312, an image recognition module 314, a loss assessment module 316, a credit review module 318, a claim payment module 320, and a loss assessment model 322. The communication module 312 is used to receive car damage image data from the car insurance claims application client 302 and transmit the data to the image recognition module 314. The image recognition module 314 is used to perform image recognition on the received car damage image. Specifically, the received car damage images are filtered first, that is, images that do not meet the requirements are filtered based on whether the automatic damage recognition requirements are met, and the car insurance claims application client 302 is notified through the communication module 312 Guide the user to take the corresponding image again. After receiving all necessary car damage images, the image recognition module 314 performs noise removal, part recognition, damage detection, cause determination, and degree determination on these car damage images. These functions will be described in more detail below, and these functions are based on the loss assessment model 322 included in the auto insurance claims application server 310. However, before using the fixed loss model 322 for image recognition, the training and learning of the fixed loss model 322 is a major challenge. In recent years, deep learning and computer vision technology have made great progress. Some simple tasks (such as ImageNet classification) have even reached a higher accuracy than humans. However, in the face of a complex real-life scene of car insurance damage determination, calculations The tough road to law is still full of difficulties, and there are still few effective technical solutions in the field of auto insurance loss determination. In real scenes, the algorithm needs to deal with a variety of interference factors such as complex lighting conditions, stains, water droplets, and vehicle structure, and learn key information that is effective for damage determination from massive data. For this reason, in one embodiment of the present invention, firstly, the tens of millions of chaotic car insurance damage history pictures from various insurance companies are structured, organized, cleaned, and marked as necessary. This huge image database is The number of photos and the complexity of tags are an order of magnitude higher than the existing ImageNet. Subsequently, the sorted, cleaned, and annotated historical pictures of car insurance damages are provided to a given damage model 322 for learning and training on a software and hardware integrated heterogeneous machine learning platform based on ASIC, FPGA, GPU and other chip technologies. The damage assessment model 322 utilizes a large amount of historical damage assessment data that has been deposited to perform model iterative learning for different models, colors and lighting conditions, and finally can output more accurate parts recognition results and target various degrees of scratching, deformation, The conclusion of damage assessment of parts such as cracking and falling off. Specifically, the damage-constrained model 322 uses a deep neural network to detect the damaged part of the vehicle and its area in the image. In an exemplary and non-limiting embodiment of the present invention, a convolutional neural network (Convolutional Neural Network, CNN) and a region proposal network (Region Proposal Network, RPN) may be combined with a pooling layer and a fully connected layer. The loss assessment model 322 is constructed in advance, and then the loss assessment model 322 is trained to generate a deep neural network through a large number of collated, cleaned and annotated auto insurance loss assessment historical pictures from various insurance companies. In other embodiments, a Fully-Connected Layer (FC), a pooling layer, a data normalization layer, etc. can also be combined. In another embodiment, if it is necessary to classify the damaged part, a probability output layer (Softmax) or the like can also be added to the damage model 322. Convolutional neural network generally refers to a neural network composed of a convolutional layer as the main structure and combined with other activation layers, which are mainly used for image recognition. The deep neural network described in this embodiment may include a convolutional layer and other important layers (such as a pooling layer, a data normalization layer, an activation layer, etc.), which are combined with a regional suggestion network (RPN) to jointly build and produce. Convolutional neural networks usually combine two-dimensional discrete convolution operations in image processing with artificial neural networks. This convolution operation can be used to automatically extract features. The Regional Proposal Network (RPN) can take the features extracted from an image (of any size) as input (the two-dimensional features extracted by the convolutional neural network can be used), and output a collection of rectangular target suggestion blocks, each block has an object Score. In another embodiment, the used convolutional neural network (CNN) may be called a convolutional layer (CNN), and the regional suggestion network (RPN) may be called a regional suggestion layer (RPN). In other embodiments of the present invention, the loss-constrained model 322 may also be combined with a modified network model based on the convolutional neural network or a modified regional suggestion network to construct the generated depth volume after training with sample data. Product neural network, namely the constant loss model 322. The models and algorithms used in the above embodiments can be selected from the same kind of models or algorithms. Specifically, for example, in the constant loss model 322, various models and variants based on convolutional neural networks and regional suggestion networks, such as Faster R-CNN, YOL0, Mask-FCN, etc., can be used. The convolutional neural network (CNN) can use any CNN model, such as ResNet, Inception, VGG, etc. and their variants. Usually the convolutional network part of the neural network can be used in mature network structures that achieve better results in object recognition, such as Inception, ResNet and other networks. In the ResNet network, the input is a picture, and the output is a plurality of picture regions containing damage parts and the corresponding damage classification (damage classification is used to determine the damage type) and confidence (confidence is the degree of authenticity of the damage type Parameter). Faster R-CNN, Y0L0, Mask-FCN, etc. are all deep neural networks including convolutional layers that can be used in this embodiment. The damage model 322 described herein can detect the damaged part, the type and degree of damage, and the location area of the damaged part in the part image by combining the region suggestion layer and the CNN layer. Returning to FIG. 3, the image recognition module 314 then transmits the recognition result to the loss assessment module 316, and the loss assessment module 316 then transmits the loss assessment result or detailed information about the loss assessment to the corresponding insurance company, that is, the user purchases it The insurance company of auto insurance, so that the insurance company can find the OE code of the damaged parts in their respective databases based on the vehicle identification code (VIN code) (obtained based on the license plate number), and then find the local repair and repair parts based on the OE code. The replacement price is combined with the damage detailed information from the image recognition module 314 to generate a corresponding repair list. The insurance company then sends the corresponding repair list back to the loss assessment module 316. The quotation of each insurance company may be different, and the final price may also be different. The loss assessment module 316 then transmits the repair list to the auto insurance claims application client 302 through the communication module 312, and the auto insurance claims application client 302 presents the repair list to the user through the UI presentation module 304. FIG. 4 shows an example of a graphical user interface 400 presented to the user by the UI presentation module 304. It can be understood that the various embodiments of the present invention are not limited to this exemplary user interface 400. As shown in the figure, the UI presentation module 304 displays the total amount of local accident damages and the repair list to the user. If the user approves the repair list, the user can activate the "accept" button 404 to continue to the next step, namely credit review step. If the user does not approve the repair list (for example, damage is not recognized, or thinks that more expensive repairs should be performed), you can choose to reshoot, that is, the user can activate the "reshoot" button 406 to return to the car damage image capture Steps for re-shooting the damaged image of the car and re-determining the damage. Or, the user can abandon the automatic claim settlement and switch to manual processing, just like the traditional auto insurance report and claim settlement process. In addition, the user can activate the "coming year premium forecast" button 402 on the right side of the total amount of fixed loss to predict the next year's premium through the auto insurance claims application based on local insurance, number of insurances this year, claim amount, personal credit and other factors. The personal credit here can be given by a third-party credit service provider. It is understood that third-party credit service providers include but are not limited to Ant Financial, and personal credit includes but not limited to the auto insurance credit developed by Ant Financial. , Which will be described in more detail below. As those skilled in the art can understand, the user interface 400 shown in FIG. 4 is only for illustrative purposes but is exemplary and non-limiting. When the user approves the maintenance list presented, that is, when the user activates the "accept" button 404, the credit review module 318 reviews the user's personal credit to determine whether to perform automatic instant compensation or transfer to manual processing. In an embodiment of the present invention, the auto insurance claims application program server 310 uses the auto insurance score developed by Ant Financial to determine the user's personal credit score, but it should be understood that other credit systems or levels may also be used, such as Ant Financial. Sesame Credit developed by Jinfu Company. Traditionally, to predict a customer’s insurance amount or loss ratio in the next year, the general practice in the auto insurance industry is to calculate based on the customer’s previous year’s insurance coverage, region, model, car price, and nature of use. factor. The car insurance score can also be based on people (such as gender, age, occupation, identity), behavior (such as the number of violations, whether to use the highway frequently, credit history, consumption habits, driving habits), and use environment (such as road type, road congestion) Three factors are used to more accurately predict the insurance amount or loss ratio of the customer in the next year, which greatly enriches the "person-related" data dimension in auto insurance. The car insurance score is based on accurate portraits of the car owner and risk analysis, for example, a score ranging from 300 to 700 is obtained. The higher the score, the lower the risk. Low-risk car owners generally have good driving habits and credit behavior. With the authorization of the user, the insurance company can query the user's auto insurance standard score, or combine its own data to process and model the label, and obtain its own auto insurance special points, thereby making more fair auto insurance pricing based on the auto insurance score. Specifically, through rapid tag mining of massive data and machine learning algorithms used to improve prediction performance, the integration of auto insurance claims data accumulated by insurance companies and user profile data accumulated by credit service providers (all after de- Sensitive processing) to generate the user’s auto insurance points. Returning to the description of the credit review module 318, when it is determined that the user’s credit is high enough, the credit review module 318 can notify the compensation module 320 to perform automatic instant compensation, that is, automatically determine the total amount of damage without manual review. Remit to the bank account or online account specified by the user. As an example and not a limitation, sufficiently high credit may mean that the credit or one of its manifestations meets or exceeds a certain threshold or level, such as a car insurance score exceeding 600 points. In an embodiment of the present invention, if the automatic payment is completed, but the car owner is found to be fraudulent in the material review stage afterwards, the insurance company will lower the car owner's credit to prevent the car owner from continuing to commit fraud. In addition, insurance companies can also cooperate with third-party neutral platforms to transfer the claim-related information to the third-party platform after desensitization that meets legal requirements for cross-insurance company credit record collection, and in multiple insurance companies To prevent car owners from making fraudulent claims in multiple insurance companies. The third-party platforms described here include but are not limited to China Insurance. If it is determined that the user's credit score is not high enough, the credit review module 318 notifies the compensation module 320 to switch the user's current claims settlement process to manual processing, just like the traditional auto insurance claims process. Fig. 5 shows another example of a car insurance claims application according to another embodiment of the present invention. In this other embodiment of the present invention, instead of taking a car damage image and sending it to the server for technical processing and then returning to the client, it can also guide the user to shoot videos around the vehicle to provide a richer vehicle image. Like information, and thanks to the significant improvement in the processor performance and image processing capabilities of smartphones in recent years, the functions of the auto insurance claims application server 310 can be integrated into the auto insurance claims application client 302 to enable the client 302 can automatically make real-time analysis of car damage. Specifically, AR (augmented reality) technology can be used to superimpose damage determination and repair information in real time when the user is shooting a video, thereby directly and instantly telling the user the degree of damage, and automatically recommending a repair list to the user, which can greatly improve shooting guidance , Timeliness of feedback, and even real-time loss determination. As shown in the diagram 502 in Figure 5, the car insurance claims application can guide the user to shoot higher quality video images in real time when the user is shooting a video, such as guiding the user to get closer to the damage site in order to make a more accurate car damage determination . In addition, as shown in the diagram 504 in FIG. 5, when the user makes a real-time damage determination when shooting a video, it can also be combined with the damage cause determination to make higher-level intelligent determinations such as "suspected damage not in this case". Moreover, since the user’s vehicle image information is not uploaded to the server, but the local CPU and GPU capabilities of the mobile computing device or tablet computing device held by the user are used, it can solve user privacy and information security to a certain extent. The problem. FIG. 6 shows a detailed functional block diagram of the image recognition module 314 according to an embodiment of the present invention. In an embodiment of the present invention, the image recognition module 314 first filters the received car damage images through the image filtering component 602, that is, filters out unqualified images based on whether the requirements for automatic damage recognition are met. , And notify the auto insurance claims application client 302 through the communication module 312 to guide the user to take the corresponding image again. For example, after the user takes a panoramic image of the vehicle and uploads it, the image filtering component 602 in the image recognition module 314 can determine whether the panoramic image includes a license plate, whether the view is too far, and so on. As another example, after the user takes a car damage detail image and uploads it, the image filtering component 602 can determine whether the car damage detail image completely includes car damage details, whether the details are clearly visible, and so on. If the uploaded image does not meet the requirements for automatic damage identification, the image filtering component 602 filters the image and guides the user to take a corresponding image again through the UI presentation module 304. After all the images necessary for automatic damage recognition have been received, the noise removal component 604 in the image recognition module 314 uses the damage model 322 to identify and exclude light reflections, shadows, reflections, stains, water droplets, and car models. Various noise interference factors such as structure. The loss assessment model 322 has been trained and learned through the accumulated historical loss assessment data (especially image data with noise), and thus can accurately identify and eliminate various interference factors. As shown in a diagram 702 in FIG. 7, the reflection noise removal based on the constant loss model 322 is shown. After the noise removal component 604 is processed, the reflection on the left front fender of the vehicle is removed. Then, the parts recognition component 606 recognizes various parts of the vehicle, including damaged parts, through the damage model 322 based on the distant view image of the damaged part. Specifically, part recognition refers to detecting the category and location of each part from an image of a car, such as identifying where in the picture is the front cover, left headlight, bumper, grille, etc. In an embodiment of the present invention, based on the loss-constant model 322 combined with Faster R-CNN (Faster Regional Convolutional Neural Network) and other network technologies to complete the part recognition, but it should be understood that in other embodiments can also be Combine other technologies to complete part identification. After identifying the various parts of the vehicle, the damage detection component 608 detects the damage location and category through the damage assessment model 322 trained with a large number of samples. Damaged parts refer to specific parts that are damaged. Damage categories include, for example, scratching, deformation, cracking, and falling off. If the shooting angle of the image including the damaged part is not good, correct the angle. Specifically, because users generally do not undergo professional training, it is naturally impossible to ensure that every uploaded photo is taken correctly and clearly. Therefore, it is sometimes necessary to correct the image to better perform damage detection, cause judgment and Judgment of damage degree. Correction techniques include, but are not limited to, projection-based methods, Hough transform, linear fitting, and Fourier transform. An illustration 704 in FIG. 7 shows a correction example of an image including a damaged part. Subsequently, the cause judgment component 610 judges the cause of the damage, such as a bicycle scratch, a sharp object damage, a two-vehicle scratch, a two-vehicle collision, etc., through the damage model 322 trained with a large number of samples. In addition, the cause judgment component 610 can also identify suspected damages that are not in this case, for example, based on the color of the damage being significantly different or the damage site does not match the collision site, and so on. Finally, the degree determination component 612 determines the degree of damage through the damage model 322 trained with a large number of samples, such as slight scratches (no primer exposed), severe scratches (exposed primer), slight deformation, severe deformation, and slight cracking. , Severe cracking, scrapping, etc., in order to generate the corresponding maintenance list. The loss constant model 322 described here has been described in detail above, so it will not be repeated here. The embodiments of the present invention are not limited to the aforementioned degree of damage. Fig. 8 shows a flow chart of a method 800 for auto insurance automatic payment according to an embodiment of the present invention. In various embodiments, the steps shown in FIG. 8 can be implemented through hardware (for example, processor, engine, memory, circuit), software (for example, operating system, application, driver, machine/processor executable instructions) Or a combination thereof. As those of ordinary skill in the art will understand, various embodiments may include more or fewer steps than shown. At 802, the user is guided to take a vehicle image and obtain the vehicle image. The UI display module in the auto insurance claim application client 302 can be used to guide the user to take car damage images through the mobile computing device or tablet computing device installed with the auto insurance claim application client 302, including the entire vehicle image containing the license plate information The image, the distant image of the damaged part, the close-up image of the damaged part, and the detailed image of the damaged car. In addition, it can also determine whether the image taken by the user meets the requirements of automatic damage recognition, and guide the user to retake the corresponding image to meet the requirements of automatic damage recognition when it is determined that the image taken by the user does not meet the requirements of automatic damage recognition. In another embodiment of the present invention, the user can also be guided to shoot a car damage video, so that more abundant car damage information can be obtained for automatic real-time damage assessment. At block 804, image recognition is performed on the acquired vehicle image to identify vehicle damage. Image recognition includes image filtering, noise removal, part recognition, damage detection, cause determination, and degree determination. These functional steps will be described in more detail in FIG. 9. In an embodiment of the present invention, these image recognition functions are accomplished through a loss-based model trained on a large number of samples based on a deep neural network. At block 806, a repair list is generated based on the vehicle damage. Vehicle damage includes damage location, damage category and degree of damage. For parts that are slightly scratched, spray paint can be used for repair, for parts that are deformed can be repaired by sheet metal, and for serious damage such as cracks and fall off, parts can be replaced directly. After sending the damage assessment result or detailed information to the corresponding insurance company, the insurance company searches for the OE code of the damaged part in their respective database based on the vehicle identification code (VIN code) (obtained based on the license plate number), and then According to the OE code, the local repair and replacement prices of parts can be found, and the corresponding repair list will be generated. At block 808, the user is asked whether to accept the generated maintenance list. If the user accepts, the flow continues to block 810, otherwise the flow returns to block 802 to re-guide the user to take the vehicle image. At block 810, the user's personal credit is reviewed. Personal credit can be combined with claims information provided by insurance companies and personal credit information provided by third-party companies. In an embodiment of the present invention, the user's personal credit may take the form of auto insurance points developed by Ant Financial, but it should be understood that other credit evaluation forms may be used. At block 812, it is determined whether the user's credit is high enough to make a determination of whether to perform automatic instant compensation or transfer to manual processing. If it is determined that the user's credit is sufficiently high, the flow continues to block 814, and the user is automatically compensated. Otherwise, the process continues to block 816, and the claim settlement is exclusively processed manually according to the traditional claim settlement process. FIG. 9 shows a flowchart of a method 900 for image recognition according to an embodiment of the present invention. In various embodiments, the steps shown in FIG. 9 can be implemented through hardware (for example, processor, engine, memory, circuit), software (for example, operating system, application, driver, machine/processor executable instructions) Or a combination thereof. As those of ordinary skill in the art will understand, various embodiments may include more or fewer steps than shown. At 902, the received car damage image is filtered. In other words, the unqualified image is filtered out based on whether the automatic damage recognition requirement is met, and the image is filtered out if it is determined that the image does not meet the requirement, and the user is guided to retake the corresponding image. At 904, noise removal is performed on the filtered car damage image. Specifically, a variety of noise interference factors such as light reflections, shadows, reflections, stains, water droplets, and vehicle structure are identified and eliminated. At 906, various parts of the vehicle are identified, including damaged parts. The logo is based on the long-range image of the damaged part, and detects the category and location of each part from the image of a car, such as identifying where the front cover, left headlight, bumper, grille, etc. . In 908, inspect the damaged part and category in the car damage image to determine the specific damaged part and the damaged category. In an embodiment of the present invention, if the shooting angle of the image including the damaged part is not good, the angle is corrected to enable better damage detection, cause judgment, and damage degree judgment. At 910, the cause of the injury is determined. Causes of damage include bicycle scratches, sharps damage, double-vehicle scratches, and double-vehicle collisions. In addition, it can also identify suspected damages that are not in this case, for example, based on the color of the damage is significantly different or the damage site does not match the collision site, etc. At 912, the degree of damage is determined in order to generate a corresponding repair list. The degree of damage includes slight scratching, severe scratching, slight deformation, severe deformation, slight cracking, severe cracking, scrapping, and so on. The above describes the embodiments of the present invention with reference to the block diagrams and/or operation instructions of the method, system and computer program product according to the embodiments of the present invention. The functions/actions indicated in the blocks can appear in an order different from that shown in any flowchart. For example, depending on the function/action involved, two blocks shown in succession may actually be executed substantially simultaneously, or the blocks may sometimes be executed in the reverse order. The above description, examples and materials provide a comprehensive description of the manufacture and use of the components of the invention. Because many embodiments of the present invention can be made without departing from the spirit and scope of the present invention, the present invention falls within the scope of the appended patent application.

100:行動計算設備 102:系統 105:觸控螢幕顯示器 110:輸入按鈕 115:側面輸入元件 120:LED 125:揚聲器 130:板載相機 135:小鍵盤 160:處理器 162:記憶體 164:作業系統 166:應用程式 168:儲存 170:電源 172:無線電介面層 174:音頻介面 176:視頻介面 26:車險理賠應用程式 202:車險理賠應用程式客戶端 204:平板計算設備 206:行動計算設備 208:網路 210:伺服器 212:車險理賠應用程式伺服器端 214:定損模型 216:儲存 218:保險公司 300:車險理賠應用程式 302:車險理賠應用程式客戶端 304:UI呈現模組 306:圖像獲取模組 308:通信模組 310:車險理賠應用程式伺服器端 312:通信模組 314:圖像識別模組 316:定損模組 318:信用審查模組 320:賠付模組 322:定損模型 400:用戶介面 402、404、406:按鈕 502、504、702、704:圖示 602:圖像過濾組件 604:雜訊去除組件 606:零件識別組件 608:損傷檢測組件 610:原因判斷組件 612:程度判定組件 800、900:方法 802~816:步驟 902~912:步驟 100: mobile computing device 102: System 105: Touch screen display 110: Enter button 115: side input element 120: LED 125: speaker 130: Onboard camera 135: Small keyboard 160: processor 162: Memory 164: Operating System 166: Application 168: Storage 170: Power 172: Radio Interface Layer 174: Audio Interface 176: Video Interface 26: Auto insurance claims application 202: Auto Insurance Claim Application Client 204: Tablet computing device 206: Mobile Computing Device 208: Network 210: server 212: Server side of auto insurance claims application 214: Fixed Loss Model 216: storage 218: Insurance Company 300: Auto Insurance Claim Application 302: Auto Insurance Claim Application Client 304: UI presentation module 306: Image acquisition module 308: Communication module 310: Server side of auto insurance claims application 312: Communication module 314: Image recognition module 316: Fixed Loss Module 318: Credit Review Module 320: Compensation Module 322: Fixed Loss Model 400: User interface 402, 404, 406: buttons 502, 504, 702, 704: icon 602: Image filtering component 604: Noise removal component 606: Part recognition component 608: Damage Detection Components 610: cause judgment component 612: Degree Determination Component 800, 900: method 802~816: steps 902~912: steps

為了能詳細理解本發明的以上陳述的特徵所用的方式，可參照各方面來對以上簡要概述的內容進行更具體的描述，其中一些方面在圖式中闡示。然而應該注意，圖式僅闡示了本發明的某些典型方面，故不應被認為限定其範圍，因為本描述可允許有其他等同有效的方面。 圖1A、圖1B和圖2示出了其中可實施本發明的各實施例的各種操作環境。 圖3示出了根據本發明的一個實施例的車險理賠應用程式的一個示例的方塊圖。 圖4示出了根據本發明的一個實施例的由UI呈現模組呈現給用戶的圖形用戶介面的示例。 圖5示出了根據本發明的另一實施例的車險理賠應用程式的另一示例。 圖6示出了根據本發明的一個實施例的圖像識別模組的詳細功能方塊圖。 圖7示出了根據本發明的一個實施例的圖像識別中的反光去除和角度矯正的示例。 圖8示出了根據本發明的一個實施例的用於車險自動賠付的方法的流程圖。 圖9示出了根據本發明的一個實施例的用於圖像識別的方法的流程圖。In order to understand in detail the manner in which the above stated features of the present invention are used, the content briefly summarized above can be described in more detail with reference to various aspects, some of which are illustrated in the drawings. It should be noted, however, that the drawings only illustrate some typical aspects of the present invention, and therefore should not be considered as limiting its scope, because the description may allow other equivalent and effective aspects. Figures 1A, 1B, and 2 show various operating environments in which various embodiments of the present invention may be implemented. Fig. 3 shows a block diagram of an example of a car insurance claims application program according to an embodiment of the present invention. Fig. 4 shows an example of a graphical user interface presented to a user by a UI presentation module according to an embodiment of the present invention. Fig. 5 shows another example of a car insurance claims application according to another embodiment of the present invention. Fig. 6 shows a detailed functional block diagram of an image recognition module according to an embodiment of the present invention. Fig. 7 shows an example of reflection removal and angle correction in image recognition according to an embodiment of the present invention. Fig. 8 shows a flowchart of a method for automatic payment of auto insurance according to an embodiment of the present invention. Fig. 9 shows a flowchart of a method for image recognition according to an embodiment of the present invention.

204:平板計算設備 204: Tablet computing device

206:行動計算設備 206: Mobile Computing Device

208:網路 208: Network

210:伺服器 210: server

212:車險理賠應用程式伺服器端 212: Server side of auto insurance claims application

214:定損模型 214: Fixed Loss Model

216:儲存 216: storage

218:保險公司 218: Insurance Company

Claims (23)

一種用於車險自動賠付的方法，該方法包括：獲取車輛圖像；對所獲取的車輛圖像進行圖像識別以標識車輛損傷；基於該車輛損傷來產生維修清單；審查用戶信用；以及自動賠付用戶，其中，該方法由包括車險理賠應用程式客戶端和車險理賠應用程式伺服器端的車險理賠應用程式來執行，並且其中，該車險理賠應用程式伺服器端與複數個保險公司對接以便從各個保險公司接收各種保險資料以及向保險公司反饋理賠決策。 A method for automatic payment of auto insurance, the method comprising: acquiring a vehicle image; performing image recognition on the acquired vehicle image to identify vehicle damage; generating a maintenance list based on the vehicle damage; reviewing user credit; and automatic payment The user, wherein the method is executed by the auto insurance claims application client and the auto insurance claims application server end, and wherein the auto insurance claims application server end is connected to a plurality of insurance companies so as to obtain information from each insurance company. The company receives various insurance materials and feeds back claims decisions to the insurance company. 如請求項1所述的方法，其中，該車輛圖像透過引導該用戶拍攝該車輛圖像來獲取。 The method according to claim 1, wherein the vehicle image is obtained by guiding the user to take the vehicle image. 如請求項1所述的方法，其中，該用戶信用在該用戶接受該維修清單的情況下審查，並且該方法還包括在該用戶不接受該維修清單的情況下重新獲取車輛圖像。 The method according to claim 1, wherein the credit of the user is reviewed when the user accepts the maintenance list, and the method further includes reacquiring the vehicle image when the user does not accept the maintenance list. 如請求項1所述的方法，其中，該自動賠付是在該用戶信用足夠高的情況下執行的，並且該方法還包括在該用戶信用不夠高的情況下將該用戶的本次理賠流程轉為人工 處理。 The method according to claim 1, wherein the automatic compensation is executed when the user's credit is sufficiently high, and the method further includes transferring the user's current compensation process when the user's credit is not high enough Artificial deal with. 如請求項1所述的方法，其中，該各種保險資料包括零配件資料、車型資料、賠付率資料、出險記錄資料、維修資料等等。 The method according to claim 1, wherein the various insurance data include parts data, vehicle model data, compensation rate data, insurance record data, maintenance data, and so on. 如請求項1所述的方法，其中，該方法由包括只包括車險理賠應用程式客戶端的車險理賠應用程式來執行。 The method according to claim 1, wherein the method is executed by an auto insurance claims application program including only an auto insurance claim application client. 如請求項1所述的方法，其中，該車輛損傷的標識也能基於該用戶拍攝的車輛視頻。 The method according to claim 1, wherein the identification of the vehicle damage can also be based on a vehicle video taken by the user. 如請求項1所述的方法，其中，該圖像識別基於定損模型，該定損模型透過深度神經網路來實現。 The method according to claim 1, wherein the image recognition is based on a constant loss model, and the constant loss model is implemented through a deep neural network. 如請求項8所述的方法，其中，該圖像識別包括圖像過濾、圖像去雜訊、零件識別、損傷檢測、原因判斷、以及程度判定。 The method according to claim 8, wherein the image recognition includes image filtering, image denoising, part recognition, damage detection, cause determination, and degree determination. 如請求項1所述的方法，其中，如果在完成該自動賠付後在審核材料階段發現該用戶有欺詐行為，則降低該用戶信用，並且將該資訊傳送到第三方平台以進行跨保險公司的信用記錄徵集並在多個保險公司之間共享。 The method according to claim 1, wherein if the user is found to have fraudulent behavior in the review material stage after the automatic payment is completed, the user’s credit is reduced, and the information is transmitted to a third-party platform for cross-insurance company Credit records are collected and shared among multiple insurance companies. 如請求項1所述的方法，其中，該用戶信用基於來自保險公司的理賠資料以及來自第三方信用服務提供商的個人信用資料兩者。 The method according to claim 1, wherein the user credit is based on both claims information from an insurance company and personal credit information from a third-party credit service provider. 一種用於車險自動賠付的系統，該系統包括：用於獲取車輛圖像的裝置；用於對所獲取的車輛圖像進行圖像識別以標識車輛損傷的裝置；用於基於該車輛損傷來產生維修清單的裝置；用於審查用戶信用的裝置；以及用於自動賠付用戶的裝置，其中，該系統包括車險理賠應用程式客戶端和車險理賠應用程式伺服器端，並且其中，該車險理賠應用程式伺服器端與複數個保險公司對接以便從各個保險公司接收各種保險資料以及向保險公司反饋理賠決策。 A system for automatic payment of car insurance. The system includes: a device for acquiring a vehicle image; a device for performing image recognition on the acquired vehicle image to identify a vehicle damage; A device for the maintenance list; a device for reviewing user credit; and a device for automatically paying users, wherein the system includes a car insurance claims application client and a car insurance claims application server, and among them, the car insurance claims application The server side is connected with multiple insurance companies to receive various insurance information from various insurance companies and feedback claims settlement decisions to the insurance companies. 如請求項12所述的系統，其中，該車輛圖像透過引導該用戶拍攝該車輛圖像來獲取。 The system according to claim 12, wherein the vehicle image is obtained by guiding the user to take the vehicle image. 如請求項12所述的系統，其中，該用戶信用在該用戶接受該維修清單的情況下審查，並且該系統還包括用於在該用戶不接受該維修清單的情況下重新獲取車輛圖像的裝置。 The system according to claim 12, wherein the credit of the user is reviewed when the user accepts the maintenance list, and the system further includes a method for reacquiring the vehicle image when the user does not accept the maintenance list Device. 如請求項12所述的系統，其中，該自動賠付是在該用戶信用足夠高的情況下執行的，並且該系統還包括用於在該用戶信用不夠高的情況下將該用戶的本次理賠流程轉為人工處理的裝置。 The system according to claim 12, wherein the automatic payment is executed when the user's credit is sufficiently high, and the system further includes a method for settling the user's current claims when the user's credit is not high enough The process is converted to a manual processing device. 如請求項12所述的系統，其中，該各種保險資料包括零配件資料、車型資料、賠付率資料、出險記錄資料、維修資料等等。 For example, the system according to claim 12, wherein the various insurance data include spare parts data, vehicle model data, compensation rate data, insurance record data, maintenance data, and so on. 如請求項12所述的系統，其中，該系統只包括車險理賠應用程式客戶端。 The system according to claim 12, wherein the system only includes an auto insurance claims application client. 如請求項12所述的系統，其中，該車輛損傷的標識也能基於該用戶拍攝的車輛視頻。 The system according to claim 12, wherein the identification of the vehicle damage can also be based on the vehicle video taken by the user. 如請求項12所述的系統，其中，用於圖像識別的裝置基於定損模型，該定損模型透過深度神經網路來實現。 The system according to claim 12, wherein the device for image recognition is based on a constant loss model, and the constant loss model is implemented through a deep neural network. 如請求項19所述的系統，其中，該用於圖像識別的裝置包括用於圖像過濾的裝置、用於圖像去雜訊的裝置、用於零件識別的裝置、用於損傷檢測的裝置、用於原因判斷的裝置、以及用於程度判定的裝置。 The system according to claim 19, wherein the device for image recognition includes a device for image filtering, a device for image denoising, a device for part recognition, and a device for damage detection. Device, device for reason determination, and device for degree determination. 如請求項12所述的系統，其中，如果在完成該自動賠付後在審核材料階段發現該用戶有欺詐行為，則將降低該用戶信用，並且將該資訊傳送到第三方平台以進行跨保險公司的信用記錄徵集並在多個保險公司之間共享。 The system according to claim 12, wherein if the user is found to have fraudulent behavior in the review material stage after the automatic payment is completed, the user’s credit will be reduced, and the information will be transmitted to a third-party platform for cross-insurance The credit history is collected and shared among multiple insurance companies. 如請求項12所述的系統，其中，該用戶信用基於來自保險公司的理賠資料以及來自第三方信用服務提供商的個人信用資料兩者。 The system according to claim 12, wherein the user credit is based on both claims information from an insurance company and personal credit information from a third-party credit service provider. 一種包含指令的電腦可讀取儲存媒體，該指令用於執行如請求項1-11中的任一項所述的方法。 A computer-readable storage medium containing instructions for executing the method described in any one of claims 1-11.
TW108136377A 2018-12-25 2019-10-08 Auto insurance automatic compensation method and system TWI710997B (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811594050.1 2018-12-25
CN201811594050.1A CN110009508A (en) 2018-12-25 2018-12-25 A kind of vehicle insurance compensates method and system automatically

Publications (2)

Publication Number Publication Date
TW202025060A TW202025060A (en) 2020-07-01
TWI710997B true TWI710997B (en) 2020-11-21

Family

ID=67165211

Family Applications (1)

Application Number Title Priority Date Filing Date
TW108136377A TWI710997B (en) 2018-12-25 2019-10-08 Auto insurance automatic compensation method and system

Country Status (3)

Country Link
CN (1) CN110009508A (en)
TW (1) TWI710997B (en)
WO (1) WO2020134943A1 (en)

Families Citing this family (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110009508A (en) * 2018-12-25 2019-07-12 阿里巴巴集团控股有限公司 A kind of vehicle insurance compensates method and system automatically
CN110443717A (en) * 2019-07-16 2019-11-12 阿里巴巴集团控股有限公司 A kind of settlement of insurance claim method and system based on credit evaluation
JP6714253B1 (en) * 2019-08-30 2020-06-24 ゼアーウィンスリーサービス株式会社 A vehicle member damage repair determination system, a device for determining whether or not a vehicle member damage warranty application is possible, an operating method of a device for determining whether a vehicle member damage warranty application is possible, and a program.
CN111461905A (en) * 2020-05-28 2020-07-28 深圳麦亚信科技股份有限公司 Vehicle insurance fraud and claim evasion method and device, computer equipment and storage medium
TWI818181B (en) * 2020-06-23 2023-10-11 新局數位科技有限公司 Car damage assessment system and implementation method thereof
CN111784685B (en) * 2020-07-17 2023-08-18 国网湖南省电力有限公司 Power transmission line defect image identification method based on cloud edge cooperative detection
CN112036283A (en) * 2020-08-25 2020-12-04 湖北经济学院 Intelligent vehicle damage assessment image identification method
CN112037235B (en) * 2020-08-27 2023-01-10 平安科技（深圳）有限公司 Injury picture automatic auditing method and device, electronic equipment and storage medium
CN112465018B (en) * 2020-11-26 2024-02-02 深源恒际科技有限公司 Intelligent screenshot method and system of vehicle video damage assessment system based on deep learning
CN112446513A (en) * 2020-12-15 2021-03-05 恒大恒驰新能源汽车研究院(上海)有限公司 Vehicle part detection method and device and computer readable storage medium
CN112712498A (en) * 2020-12-25 2021-04-27 北京百度网讯科技有限公司 Vehicle damage assessment method and device executed by mobile terminal, mobile terminal and medium
CN112733635A (en) * 2020-12-28 2021-04-30 北京迈格威科技有限公司 Object identification method and device and electronic equipment
CN114529420B (en) * 2022-03-08 2023-02-28 上海恒格信息科技有限公司 Intelligent guarantee expert system based on insurance demand
CN115511124B (en) * 2022-09-27 2023-04-18 上海网商电子商务有限公司 Customer grading method based on after-sale maintenance records
CN115410174B (en) * 2022-11-01 2023-05-23 之江实验室 Two-stage vehicle insurance anti-fraud image acquisition quality inspection method, device and system
CN115631002B (en) * 2022-12-08 2023-11-17 邦邦汽车销售服务(北京)有限公司 Computer vision-based intelligent damage assessment method and system for vehicle insurance
CN115810134B (en) * 2023-02-14 2023-07-21 之江实验室 Image acquisition quality inspection method, system and device for vehicle insurance anti-fraud
CN116935293B (en) * 2023-09-15 2024-01-02 国任财产保险股份有限公司 Automatic processing method and system for remote video exploration and damage assessment

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI465092B (en) * 2009-03-30 2014-12-11 Yuh-Shen Song Privacy protected anti-identity theft and payment network
CN106780048A (en) * 2016-11-28 2017-05-31 中国平安财产保险股份有限公司 A kind of self-service Claims Resolution method of intelligent vehicle insurance, self-service Claims Resolution apparatus and system
CN107292528A (en) * 2017-06-30 2017-10-24 阿里巴巴集团控股有限公司 Vehicle insurance Risk Forecast Method, device and server
CN108280769A (en) * 2018-02-01 2018-07-13 阿里巴巴集团控股有限公司 The data processing method of Claims Resolution business, device, electronic equipment, server
CN108647712A (en) * 2018-05-08 2018-10-12 阿里巴巴集团控股有限公司 Processing method, processing equipment, client and the server of vehicle damage identification
CN108898504A (en) * 2018-07-09 2018-11-27 北京精友世纪软件技术有限公司 The intelligent training and improving method of loss assessment system are surveyed in a kind of movement

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100145734A1 (en) * 2007-11-28 2010-06-10 Manuel Becerra Automated claims processing system
CN104268783B (en) * 2014-05-30 2018-10-26 翱特信息系统（中国）有限公司 The method, apparatus and terminal device of car damage identification appraisal
CN107679996A (en) * 2017-10-10 2018-02-09 民太安财产保险公估股份有限公司 Vehicle insurance Claims Resolution method and system
CN108009834B (en) * 2017-12-27 2021-10-15 上海唯链信息科技有限公司 Automobile insurance information system based on block chain technology
CN110009508A (en) * 2018-12-25 2019-07-12 阿里巴巴集团控股有限公司 A kind of vehicle insurance compensates method and system automatically

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI465092B (en) * 2009-03-30 2014-12-11 Yuh-Shen Song Privacy protected anti-identity theft and payment network
CN106780048A (en) * 2016-11-28 2017-05-31 中国平安财产保险股份有限公司 A kind of self-service Claims Resolution method of intelligent vehicle insurance, self-service Claims Resolution apparatus and system
CN107292528A (en) * 2017-06-30 2017-10-24 阿里巴巴集团控股有限公司 Vehicle insurance Risk Forecast Method, device and server
CN108280769A (en) * 2018-02-01 2018-07-13 阿里巴巴集团控股有限公司 The data processing method of Claims Resolution business, device, electronic equipment, server
CN108647712A (en) * 2018-05-08 2018-10-12 阿里巴巴集团控股有限公司 Processing method, processing equipment, client and the server of vehicle damage identification
CN108898504A (en) * 2018-07-09 2018-11-27 北京精友世纪软件技术有限公司 The intelligent training and improving method of loss assessment system are surveyed in a kind of movement

Also Published As

Publication number Publication date
CN110009508A (en) 2019-07-12
TW202025060A (en) 2020-07-01
WO2020134943A1 (en) 2020-07-02

Similar Documents

Publication Publication Date Title
TWI710997B (en) Auto insurance automatic compensation method and system
US11580605B2 (en) Feedback loop in mobile damage assessment and claims processing
US10803532B1 (en) Processing insured items holistically with mobile damage assessment and claims processing
US11625791B1 (en) Feedback loop in mobile damage assessment and claims processing
US11361385B2 (en) Application facilitated claims damage estimation
US20220284517A1 (en) Automobile Monitoring Systems and Methods for Detecting Damage and Other Conditions
US10573012B1 (en) Three dimensional image scan for vehicle
US20240037671A1 (en) Agent-facilitated claims damage estimation
US11915321B2 (en) Configuration and transfer of image data using a mobile device
US10572944B1 (en) Claims damage estimation using enhanced display
JP2023524609A (en) How to Determine Paint Requirements for Damaged Vehicles
US11455691B2 (en) Processing insured items holistically with mobile damage assessment and claims processing
US11380148B2 (en) 3D vehicle model data capturing and retrieving for vehicle inspection, service and maintenance
US20230020642A1 (en) Processing insured items holistically with mobile damage assessment and claims processing
US20190102749A1 (en) Methods and systems of retail automotive check-in with a mobile device
CN114663871A (en) Image recognition method, training method, device, system and storage medium
CN113065908A (en) Rental method, rental device, rental platform, and storage medium