Abstarct: Sign language recognition is the first step towards interaction between deaf community and computer systems. Although there are many previous works on American Sign Language and languages other than Persian, there are only limited efforts on Persian sign language (PSL) recognition. In this research, a robust Persian sign language recognition system baxsed on Baghcheban notation (the first and most practical PSL used) is presented. Depth cameras, geometric features of the hand, skeleton and face are extracted and the results show a high rate of recognition in an environment with complex background. This is the first research on PSL baxsed on complex backgrounds. Also this is the first time that skeleton and mouth features are used in PSL recognition. Main contributions of the proposed system include creating the first depth-baxsed Baghcheban notation PSL databaxse with full body and facial images, recognition of all PSL alphabet containing 38 letters, presenting a real-time recognition system, using skeleton and facial features in recognizing PSL and further analysis of results to achieve the maximum recognition rate of %99. The databaxse created in this research consists of 150 to 500 samples per 38 alphabet letters in a complex scene with occlusion between hand, face, body, cloth and background. Kinect sensor is used to record the RGB image, depth image, upper body skeleton and 121 facial points of the user. The results indicate a %99.4 recognition rate baxsed on geometric features of the hand and mouth, and skeleton features of the relational placement of the hand on 3800 samples and %99.2 on 5700 samples. The performance analysis on the proposed system shows a real-time speed of 40-60 fps using normal Bayes classifier.