智能化矛盾调解终端采集数据集
收藏资源简介：
通过线下方式，基于智能化矛盾调解终端在合肥市公安局庐阳分局逍遥津派出所矛盾调解室与合肥市滨湖新区矛盾调解中心进行现场采集数据。通过分析任务书考核指标，在线下通过智能化矛盾调解终端、语音拾取器、4K高清摄像头对矛盾调解中心内的当事人脸部表情、语音信息进行采集。其中采集过程保障隐私安全问题，在调解开始前获得当事人同意。自2022年11月-2024年07月已经完成全线下数据集500余例涉及家庭矛盾纠纷、劳资纠纷等多类矛盾事件。对数据类型进行分类，大致分为用户基础信息、生产数据。通过摄像头采集到的面部视频，并对视频进行分帧处理，形成单一调解事件不低于30张面部表情，并对表情进行心率变异性（HRV）数据分析处理。通过语音拾取器进行采集当事人的语音信息，并将该语音信息上传到公有云商业版语音转文字SDK（百度云），反转的文字信息自动存储在服务端，并将服务端的文字信息与矛盾调解心理矛盾纠纷干预服务平台进行数据校验，判断是否为威胁语言、报复性语言等，如果发现文字内容出现威胁性词语，系统将第一时间进行反馈预警。
Data was collected on-site via intelligent conflict mediation terminals at the Conflict Mediation Room of Xiaoyaojin Police Station, Luyang Branch, Hefei Public Security Bureau, and the Conflict Mediation Center of Binhu New Area, Hefei City, through offline methods. Based on the analysis of assessment indicators in the task document, facial expressions and voice information of the parties involved in the conflict mediation center were collected via intelligent conflict mediation terminals, voice pickups, and 4K high-definition cameras during offline mediation sessions. Privacy security was ensured throughout the collection process, with consent obtained from all parties prior to the start of mediation. From November 2022 to July 2024, over 500 complete offline dataset cases have been collected, covering various types of conflict incidents including family disputes and labor disputes. The collected data is roughly categorized into two types: user basic information and production data. Facial videos collected by the cameras were frame-split to generate no fewer than 30 facial expression frames per single mediation event, and Heart Rate Variability (HRV) data analysis was performed on the extracted facial expressions. Voice information of the parties was collected via voice pickups, then uploaded to the commercial public cloud speech-to-text SDK (Baidu Cloud). The converted text information was automatically stored on the server, and data verification was conducted between the server-stored text and the Conflict Mediation and Psychological Dispute Intervention Service Platform to determine whether the content contained threatening or retaliatory language. If threatening words were detected in the text, the system would immediately issue a feedback warning.
