Emotion recognition especially negative emotion detection has become a topic of considerable concern in both scientific research and practical applications. The inherent limitation is that a number of physiological signals are difficult to be monitored in daily life activities. This paper presents a novel method on negative emotion detection via feature fusion from only one-channel electrocardiogram (ECG) signal, which is able to instantaneously assess the subject's state even in real-time events. A series of features have been calculated from the original ECG signal and its derived heart rate variability (HRV), including linear-derived features, nonlinear-derived features, time domain (TD) features, and time-frequency domain (T-F D) features, which are then fused for classification using SVM. The new method was implemented on the Bio Vid Emo DB dataset for evaluation, where the highest accuracy of 79.51% was achieved with minor time cost of 0.13ms in the classification of positive and negative emotion states. It exhibited a better performance than the relevant studies in the comparison experiments. This method is applicable for wearable negative emotion detection due to its acceptable accuracy, real-time performance, as well as the convenience of wearable one-channel ECG acquisition in daily activities.