Emotion recognition has various applications closely linked with people's daily life, such as human-computer interaction and psychotherapy. So this paper focuses on the research of emotion recognition of body movement and proposes a novel emotion recognition approach based on weighted kernel support vector machine (SVM) using wearable inertial sensors. Specifically, the mapping relationship from emotions to body movements is established by fuzzy comprehensive evaluation. The subjects wear inertial sensors on their arms and wrists to collect data in six emotions including sleepy, bored, excited, tense, anger and distressed. In recognition phase, the weighted kernel SVM model is constructed, in which the fuzzy function is auxiliary to improve the weight calculation method of kernel functions in multiple kernel SVM. It also explores the effect of different combinations of inertial sensors on the recognition. The results show that compared with other methods, the proposed method achieves 98.4% accuracy for six emotions, which is effective and applicable.