【作 者】：
陈靓影， 张如静， 王 帆， 张 坤， 王广帅， 王广海
【关 键 词 】：
孤独症儿童； 早期社交技能； 应答性共同注意； 应答性社交微笑； 智能化评估
【栏 目】：
中小学电教
【中文摘要】：
人工智能技术的发展给孤独症评估带来了新思路，目前智能化评估集中为孤独症判别二分类评估，缺乏高效、量化的精细化评估方法。研究设计开发了一款基于人工智能技术的早期社交技能量化评估系统。实验招募了28名孤独症儿童为研究对象，采集了社交技能评估视频，记录了评估过程中儿童的头部转向活动和面部表情。使用头部姿态估计、表情识别以及语音识别等技术分析评估过程中的应答性共同注意和应答性社交微笑。结果表明：（1）早期社交技能的智能化评估得分以精确的反应时间和持续时间区分呈梯度分布，评估结果客观精细；（2）准确性检验证明智能化评估得分与人工评分结果具有一致性，验证了孤独症儿童早期社交技能智能化评估的可行性。这种评估方式客观、准确、易用，为儿童早期社交技能连续性评估提供技术支持，客观量化的精细化评估结果为孤独症儿童的精准干预提供客观依据，具有很好的应用价值和现实意义。
【英文摘要】：
The development of artificial intelligence brings new ideas to the assessment of Autism Spectrum Disorder (ASD) while most of intelligent assessments use ASD screening methods, lacking efficient and quantitative refined assessment methods. Thus, a quantitative evaluation system for early social skills based on artificial intelligence is developed. 28 children with ASD are recruited in the experiment, and the videos for the assessment of social skills are collected, in which children's head positions and facial expression are recorded. Techniques such as head pose estimation, facial expression recognition and automatic speech recognition are used to assess their social skills in response to joint attention and responsive social smile. The results indicate that (1) the score of intelligent assessment of early social skills is distributed in gradient with precise response time and duration, and the evaluation result is objective and fine. (2) Accuracy test proves that the score of intelligent assessment is consistent with that of human-rating, which verifies the feasibility of intelligent assessment of early social skills of children with ASD. This method is objective, accurate and easy to use, providing technical support for the continuous assessment of children's early social skills. The precise assessment results, which are objective and quantitative, provide an objective basis for accurate intervention of children with autism, which are of great value for application and practical significance.