CAJ | 학술논문

针对不同光照条件下基于摄像头的人体手势识别所存在的识别率、实时性和鲁棒性等问题，提出一种基于不同光照条件的人体手势识别新方法。首先利用传统图像处理技术与光线补偿、色彩平衡、对比度受限自适应直方图均衡( CLAHE)进行视频帧预处理，然后运用基于Haar分类器的前期检测与基于YCrCb肤色模型的后期验证进行手势识别分类，并提取质心特征进行方向跟踪判断，最终完成静态与动态实时人体手势识别任务。实验结果表明，在几种不同光照条件下，实时识别率达90%以上，且在手势扭曲、侧对与旋转一定程度下仍具良好识别效果。
침대불동광조조건하기우섭상두적인체수세식별소존재적식별솔、실시성화로봉성등문제，제출일충기우불동광조조건적인체수세식별신방법。수선이용전통도상처리기술여광선보상、색채평형、대비도수한자괄응직방도균형( CLAHE)진행시빈정예처리，연후운용기우Haar분류기적전기검측여기우YCrCb부색모형적후기험증진행수세식별분류，병제취질심특정진행방향근종판단，최종완성정태여동태실시인체수세식별임무。실험결과표명，재궤충불동광조조건하，실시식별솔체90%이상，차재수세뉴곡、측대여선전일정정도하잉구량호식별효과。
In order to improve the recognition rate, real-time performance and robustness of human gesture recognition under different light conditions based on camera, a new human gesture recognition approach based on different light conditions was proposed. Traditional image processing technique, light compensation, color balance and Contrast Limited Adaptive Histogram Equalization ( CLAHE) methods were applied to deal with the pre-processing of the video frame; earlier detection based on Haar classifier and post verification based on YCrCb skin color model were used for gesture recognition and classification; then the centroid features were extracted and tracked to determine direction; finally, the task of static and dynamic real-time human gesture recognition was achieved. The experimental result show that real-time recognition rate is above 90% under condition of several different lights, and good recognition effect is maintained with twisted, sided and rotated gestures to some extent.