ROBUST HAND TRACKING IN LOW-RESOLUTION VIDEO SEQUENCES

摘要

Automatic detection and tracking of human hands in video imagery has many applications. While some success has been achieved in human-computer interaction applications,hand tracking would also be extremely useful in security systems, where it could help the system to understand and predict human actions, intentions, and goals. We have designed and implemented a prototype hand tracking system,which is able to track the hands of moving humans in low resolution video sequences. Our system uses grayscale appearance and skin color information to classify image subwindows as either containing or not containing a human hand. The prototype’s performance is quite promising, detecting nearly all visible hands in a test sequence with a relatively low error rate. In future work we plan to improve the prototype and explore methods for interpreting human gestures in surveillance video.