Abstract: A kinematic information acquisition system of human lower limbs based on attitude and heading reference system (AHRS) was developed for a lower limb soft assistance robot. The AHRS of the system has the characteristics of small volume and low power consumption, and can be worn for a long time and easy to keep a tight connection with human body through the bind. The design of AHRS integrates Kalman filter (KF) intend to improve the accuracy of the data, multi-channel data communication and transmission to PC through the wireless network formed by ZigBee. Then, the kinematic model of the human lower limbs was established, and the conversion relationship between the AHRS measurement angle and the human lower limb posture angle was introduced. At the same time, considering the robot provides different forces under the different gait activities, the decision tree classifier for classification that can identify different gait activities and assist robots provide power to human effectively was designed. In a variety of gait activities, making data collected by this system to compare with data from the Vicon system shows that the data of the system has high precision, and also verifies that the system has good stability and reliability. Additionally, a large number of gait data were used to train decision tree classifier and the accuracy of gait recognition is verified by the classification experiment of unknow gait activities.