研发实力
广州广电运通智能科技有限公司是一家以人工智能技术为核心，集研发、设计、生产、销售于一身的国家高新技术企业
国家级科研平台
省级科研平台
国家重点研发计划项目
全国性科技奖
工业设计奖
参编并发布实施标准
企业荣誉
产品荣誉
知识产权
研发团队
AI algorithm capabilities
- 多模态掌纹和掌纹识别
- 3D视觉通行逻辑检测
- 车辆特征结构化分析
- 高速公路投卡识别
- 人脸结构解析
- X光机液体识别
- 人包关联分析
- 智慧客服大模型
- 行政问询大模型
- 单点自适应信号灯优化算法
- 干道绿波优化算法
- 注界数据标注平台
Vision Algorithms
Natural Language Processing Algorithms
Signal Processing Algorithms
Operational Optimization Algorithms
System Platform
Vision Algorithms
| Algorithm | Technical Introduction | Product Features | Product Advantages | Application Areas |
|---|---|---|---|---|
| Multimodal palm print and palm vein recognition | The multimodal palm print and palm vein recognition algorithm performs palm detection, key point detection, quality analysis, ROI area acquisition, feature extraction, etc. on the real-time captured palm images, combined with multimodal information fusion technology to achieve multimodal palm print and palm vein identity recognition. It can be applied in payment, attendance, finance and other scenarios. | 1. Palm ROI acquisition: palm detection and key point detection are performed on the collected palm images, and the palm ROI area is obtained by cropping. 2. Palm quality judgment: the brightness, clarity, posture angle and occlusion degree of the palm image can be judged to obtain high-quality images. 3. Palm vein recognition: high-precision identity authentication and recognition are achieved by extracting, enhancing, analyzing and comparing the distribution characteristics of human palm veins. 4. Palm print recognition: reliable individual recognition and identity authentication are achieved by extracting the texture characteristics of human palm skin, enhancing, analyzing and comparing them. 5. Feature fusion: IR+RGB image acquisition technology is used to extract rich palm features, and the palm print and palm vein fusion recognition algorithm is combined to maintain an extremely low error recognition rate in a database of tens of millions. |
1. High accuracy: Adopt feature fusion technology to maintain extremely low false recognition rate. 2. High security: Ensure that the collected information is authentic and reliable, and effectively prevent fake attacks. 3. Rapidity: Quickly complete authentication and identification, reduce waiting time in queues, and improve overall recognition efficiency. 4. Contactless identification: Achieve completely contactless authentication, convenient and hygienic. 5. Strong privacy protection: User-active identification, no need to worry about privacy leakage and false triggering of identification problems. |
|
| 3D vision pass logic detection | The 3D visual passage logic module senses and identifies the pedestrian passage status through visual perception technology, aiming to realize real-time intelligent recognition of people and objects in the gate area, accurately detect height, effectively track pedestrian ID information, and judge abnormal behaviors such as pedestrian tailing, so as to shorten and optimize the gate and improve the user's passage experience. | 1. Intelligent detection: The algorithm accurately identifies pedestrians, gates, luggage and other related information in the gate, and realizes automatic framing of the gate. 2. Target tracking: Accurately lock pedestrian ID information and determine the pedestrian's passage status. 3. Height detection: Use 3D vision technology to calculate the height of pedestrians and distinguish between adults and children. 4. Occlusion detection: Perform channel occlusion detection to reduce the risk of passengers, packages, etc. being pinched. 5. Behavior analysis: Realize the detection and analysis of behaviors such as close-range side-by-side tailing, front-and-back tailing, intrusion, and reverse intrusion. |
1. Strong biometric recognition capability: accurately distinguish between people and objects, adults and children. 2. Strong channel detection capability: optimize channel blocking strategy, greatly reduce false alarms and mis-clip situations. 3. Strong tailing detection capability: accurately detect side-by-side tailing, front-to-back tailing behaviors. 4. Low maintenance cost: can reduce the size of the gate, reduce the floor space, reduce the design of sensors and cables, and easy maintenance. |
|
| Structural analysis of vehicle characteristics | Vehicle feature structured analysis integrates multiple image processing technologies to perform vehicle model recognition, license plate recognition, body attribute analysis, vehicle line crossing detection, vehicle reversal detection, etc. It can distinguish more than a dozen types of vehicles, such as motor vehicles and non-motor vehicles, and output information such as the location, license plate number and body color attributes of the vehicle model, to achieve intelligent traffic management. | 1. Vehicle type recognition: The algorithm analyzes the characteristics and appearance of the vehicle to accurately identify various types of vehicles in various weather conditions (sunny, rainy, foggy, cloudy, etc.), different scenes, and various time periods during the day/night. 2. License plate recognition: The license plate recognition function is realized through algorithms such as license plate detection, character segmentation, and character recognition. 3. Vehicle body attribute analysis: Accurately identify the color of the vehicle, such as red, blue, and white. 4. Vehicle line crossing detection: According to the characteristics of the lane line, the position relationship between the vehicle and the lane line is analyzed to realize vehicle line crossing detection. 5. Vehicle reverse detection: Determine the driving status of the vehicle and whether it is reverse based on the vehicle position and movement trajectory. |
1. High accuracy: Based on advanced image processing and deep learning algorithms, high-accuracy recognition is achieved. 2. Real-time monitoring: It can monitor vehicle information on the road in real time and provide data support for traffic management and decision-making. 3. Efficient identification: It can quickly and accurately identify vehicle color, model, license plate and other features to achieve intelligent management. 4. Automated management: It can analyze and record vehicle information, reduce the need for manual intervention, and improve management efficiency. |
|
| Highway card identification | The highway card insertion recognition algorithm accurately identifies the card inserted by the driver when he exits the station, completes the vehicle payment card recovery process, ensures information security, and improves toll collection efficiency. | 1. Card identification: The camera-based card recognition algorithm assists the gate card collection robot on the highway to complete the automatic recovery of CPC cards, thereby improving vehicle passing efficiency. | 1. High accuracy: The algorithm accurately recognizes the card being thrown. 2. High adaptability: The algorithm effectively copes with complex situations such as lighting changes and background interference. 3. High timeliness: The algorithm module is optimized to enable fast detection and recognition in real-time video streams. |
|
| Face structure analysis | The facial structure analysis algorithm needs to be combined with the camera for video analysis and use technical means such as deep learning to achieve functions such as face detection, face tracking, face quality judgment, liveness detection, and face attribute analysis to obtain high-quality face capture. | 1. Face detection: Analyze the input image to obtain the location of the face, and identify ordinary faces and faces wearing mask. 2. Face tracking: Track the face in the image in real time. 3. Liveness detection: Determine whether the captured face is the user's real face, and avoid forged face attacks such as face pictures printed on color paper, digital face images on electronic device screens, and masks. 4. Face light intensity detection: Calculate and judge whether the light of the face image is normal, overexposed, or too dark. 5. Face clarity detection: Calculate and judge the clarity of the face image, such as normal, generally blurred, and very blurred. 6. Face posture estimation: The algorithm calculates and judges the posture angles (roll angle, pitch angle, and yaw angle) in three directions in the face space. 7. Face occlusion detection: The algorithm identifies the degree of occlusion of the facial features. |
1. High-quality capture: multiple indicators are used to judge and achieve high-quality face capture. 2. High adaptability: it can adapt to multiple cameras such as monocular, binocular, and 3D structured light. 3. High robustness: it has high accuracy and high robustness in complex environments such as different lighting. 4. High security: it has liveness detection function and can prevent various attacks. |
|
| X-ray machine liquid identification | The X-ray machine liquid recognition algorithm is mainly used to detect items carried by passengers, detect dangerous liquids/safe liquids, detect the volume of liquid containers, and reduce the rate of staff opening bags. | 1. Safe liquid identification: accurately identify common safe liquids, such as pure water, carbonated drinks, juice, etc., reduce the rate of staff opening packages, and reduce traffic pressure. 2. Hazardous liquid identification: accurately identify various types of hazardous liquids, including flammable liquids, toxic liquids, etc., effectively prevent the carrying of hazardous liquids and the occurrence of potential threats, and provide passengers with a safer and more convenient travel experience. 3. Liquid container volume detection: detect the volume of liquid containers and assist in liquid identification detection. |
1. High efficiency: It can scan and analyze luggage in a few seconds and give results quickly. 2. High accuracy: It can accurately identify the liquid state of various packages and containers. 3. Convenience: It reduces the rate of staff opening packages and facilitates problem tracing. |
|
| Human-package association analysis | Human-package association analysis includes pedestrian detection, package detection, and ID association algorithm, which can effectively identify pedestrians and their backpacks, realize human-package association, and provide better security inspection services and management. | 1. Target detection: Use deep learning and other technologies to extract pedestrian and package features to accurately detect pedestrians and packages. 2. Target tracking: ID matching and tracking of pedestrians and packages. 3. Person-package association: Associate pedestrian and package passenger information at the entrance, establish a person-package association record, and match the baggage video information with the passenger information again at the exit to achieve full monitoring of inbound and outbound baggage, reducing the phenomenon of pedestrians missing or taking the wrong baggage. |
1. High accuracy: The algorithm introduces a new spatial distance discrimination mechanism to ensure the accuracy of association analysis. 2. High stability: Special optimization is performed for common packages such as backpacks to improve the stability of detection and association. 3. High timeliness: It can perform fast detection and association in real-time video streams. |
|
Natural Language Processing Algorithms
| Algorithm | Technical Introduction | Product Features | Product Advantages | Application Areas |
|---|---|---|---|---|
| Smart Customer Service Model | The intelligent customer service system combines advanced technologies such as natural language processing, machine learning and deep learning, including algorithms such as intent recognition and text error correction. It has common dialogue functions such as route planning, fare inquiry, and line transfer inquiry, which helps alleviate the service pressure of subway staff, improve passenger experience, and achieve smarter and more efficient services. | 1. Intent recognition: Identify the user's intent based on the query information, perform corresponding post-processing steps, such as retrieving relevant information, performing corresponding operations, and providing corresponding responses or feedback to the user. 2. Entity recognition: By learning and utilizing various domain information, more accurate and comprehensive entity recognition is achieved, allowing users to obtain entity information from it. 3. User semantic analysis: The system can use the association information between intent and entity to further analyze and process the query information and obtain deeper semantic information. 4. Text error correction: Automatically detect and correct spelling, grammar, punctuation and other errors in the text to improve the accuracy and readability of the text. |
1. High accuracy: By optimizing the algorithm, the user's inquiry intention can be accurately captured and feedback can be given. 2. High timeliness: Tens of thousands of questions can be answered in real time, reducing customer service response time and greatly improving service efficiency. 3. Intelligent error correction technology: Aiming at the industry-specific business vocabulary, solve the problem that voice recognition cannot recognize professional vocabulary, self-developed industry-specific error correction algorithm, and realize intelligent error correction technology. 4. Industry-specific knowledge base: In-depth research on industry customer service content, in-depth understanding of industry-specific knowledge such as station information, station navigation facilities, and build an industry-specific knowledge base. |
|
| Administrative Inquiry Model | The administrative inquiry system is used by internal employees of the company to consult questions and answer common administrative questions. Different from traditional inquiries, it provides better user experience, smarter answers, and stronger reasoning, table analysis and summary capabilities. | 1. Intent recognition: Identify the user's intention based on the inquiry information, determine whether it is an administrative inquiry or a free inquiry, and perform corresponding post-processing steps, such as retrieving relevant information, performing corresponding operations, and providing corresponding responses or feedback to the user. 2. User semantic analysis: The system can use the associated information such as intent and documents to further analyze and process the inquiry information and obtain deeper semantic information. 3. Text error correction: Automatically detect and correct spelling, grammar, punctuation and other errors in the text to improve the accuracy and readability of the text. 4. Stop as you ask: The system supports real-time stop, and the user can end the conversation at any time. 5. Document generation: Corresponding documents can be generated based on keywords, such as meeting minutes. |
1. Support flexible Q&A: It can generate answers that conform to grammatical and semantic specifications based on the questions provided by the user. 2. High adaptability: It can adapt to the ever-changing knowledge needs and improve the accuracy and comprehensiveness of answers. 3. Support streaming output: The system can output the answer content step by step in sequence, and also supports real-time stopping. Users can end the conversation at any time. |
Company Management Public Services |
Signal Processing Algorithms
| Algorithm | Technical Introduction | Product Features | Product Advantages | Application Areas |
|---|---|---|---|---|
| Security door object detection analysis | The security gate object detection and analysis algorithm detects passing passengers and analyzes whether they are carrying prohibited items/dangerous goods/illegal items, thereby improving the security prevention capabilities of the security gate. | 1. Item detection: By detecting and analyzing the items carried by pedestrians, it is determined whether they are carrying electronic products, metal products, etc. | 1. High accuracy: It can accurately locate and identify dangerous goods and contraband, improving the detection accuracy of the security door system. 2. High adaptability: It can adapt to different security door equipment and scene requirements. 3. High efficiency: The training method is fast and stable, which greatly shortens the algorithm development cycle and improves the system's online speed and deployment efficiency. |
|
Operational Optimization Algorithms
| Algorithm | Technical Introduction | Product Features | Product Advantages | Application Areas |
|---|---|---|---|---|
| Single-point adaptive traffic light optimization algorithm | The single-point adaptive traffic light optimization algorithm aims to adaptively time the traffic lights in all directions of the intersection based on the historical road intersection data of a single intersection using a swarm intelligence optimization algorithm, thereby minimizing the waiting time of vehicles, improving intersection traffic efficiency and reducing traffic congestion. | 1. Traffic light optimization: realize adaptive timing of traffic lights at intersections, minimize waiting time for vehicles, improve traffic efficiency at intersections, and reduce traffic congestion. | 1. Adaptive adjustment of signal timing: ensure that the waiting time of vehicles at the intersection is minimized, reducing congestion and queue length. 2. Low cost consumption: no need for large-scale data collection and complex equipment requirements. 3. High robustness: It has strong robustness and adaptability and can cope with complex and changing road conditions. |
|
| Green wave optimization algorithm for main roads | The adaptive arterial green wave coordination algorithm is an important measure to ensure smooth urban traffic, which enables vehicles on the arterial road to obtain the right of way without stopping as much as possible. The algorithm consists of dynamic sub-area division, arterial green wave optimization, and adaptive single-point optimization. By detecting and analyzing real-time data within and outside the domain, it realizes real-time and accurate adaptive adjustment of intersections, which is suitable for various traffic conditions such as peak and flat peak. | 1. Dynamic sub-area division: Through dynamic data inside and outside the domain such as radar vision and the Internet, the sub-area is dynamically adjusted based on comprehensive indicators such as traffic flow steering and intersection spacing, and the division effect can be quantitatively evaluated. 2. Optimization calculation of green waves on trunk roads: Automatically obtain the optional set of phase sequence, set the two-way red and green wave schemes with common cycles and unequal cycles, and automatically adjust the one-way and two-way red and green wave schemes according to the flow changes during peak and flat peaks in the morning and evening. |
1. High efficiency and real-time performance: The endpoint intersection adopts single-point adaptive control to quickly and effectively sense the lane traffic conditions and adjust the release time of the phase in real time. 2. High reliability: The algorithm results can be traced back, and all solutions have corresponding quantitative indicators to review and evaluate the solutions before they are issued. 3. Strong interpretability: The solution calculation process is completed by a large number of mathematical derivations, and the decision-making process such as principles and optimization operators is persuasive, which can achieve accurate, fair and effective output of signal control solutions. 4. High controllability and safety: The green wave and sub-area division schemes can be viewed through the platform, and traffic control personnel can confirm the reliability of the newly generated scheme indicators to ensure that the real-time scheme of the intersection is safe and controllable. |
|
System Platform
| Algorithm | Technical Introduction | Product Features | Product Advantages | Application Areas |
|---|---|---|---|---|
| Annotation data annotation platform | Annotation Data Annotation Platform is a data annotation platform for large-scale images and videos. It has multiple functions such as project management, data management, user management, task management, data annotation, etc., providing high-quality data for the training of machine learning and deep learning models. | 1. Project management: It has management functions such as project creation, modification, and deletion. 2. Data management: It can upload, download, delete data, support pre-labeled data upload, and support multiple label formats. 3. User management: It can create user accounts, manage permissions, modify passwords, and other user account management operations. 4. Task management: It can assign tasks such as data labeling and data quality inspection, and has data return and data acceptance functions, which is convenient for tracking the status of each labeling task and is suitable for team collaboration. 5. Progress management: It has project progress display and labeling progress display functions to realize visual labeling progress management. 6. Data labeling: It supports multiple labeling methods such as rectangular labeling, point labeling, polygon labeling, ellipse labeling, etc. to meet various complex scene requirements. |
1. Ease of use: In the form of a web interface, the login method is highly convenient and does not require a complicated installation process. 2. Flexibility: Supports multiple annotation types and custom tools, and is applicable to a wide range of scenarios. 3. Intelligent assistance: It can support pre-annotation, integration of annotation and quality inspection, and accelerate the annotation progress. |
|
机电一体化能力
- 通用基础技术研究，智能伺服驱动、鸿蒙适配
- 核心部件单板设计，自主可控;
- 嵌入式软硬件的自主研发能力，敏捷开发定制
工业设计能力
- 超20年的工业设计，丰富的经验把客户需求转化为 产品设计语言能力
- 人机工程的多层次模拟与应用把握
- 对行业工业用料的精准理解与适用性引导
软件研发能力
- 基于k8s的云原生微服务架构构建城轨业务软件 的基础技术开发底座
- 业务软件的SaaS化封装，实现业务软件的客户 自定义应用
- 应用CI/CD持续集成、交付技术完成软件全生命 周期持续迭代
- 全场景设备软件：具备地铁、高铁、城际、机场、数智应用、文娱、等全场景终端设备软件开发能力，例如闸机、自助取票/验票、智能登机、自助充值、自助兑换、存款验钞、信息查询等终端设备与应用软件