In previous work, we proposed a promising method, named HWBST, for Change-Point (CP) detection from time series. However, the performance of HWBST is affected partly by the search criteria in terms of Binary Search Tree (BST). In this paper, we propose an improved method for fast CP detection from large-scale ECG time series, based on multi-level Haar Wavelet and Ternary Search Tree (HWTST, for short). In this method, we construct a ternary search tree termed TSTcD from a diagnosed time series by using multi-level HW. Then, we implement fast detection abrupt change from root to leaf nodes in TSTcD, by introducing two search criteria in terms of the data fluctuation in the left, right, and virtual middle branches of TSTcD. Based on the assembled and abnormal ECG samples, we evaluate the proposed HWTST by comparing with HWBST, KS, and T methods. The results show that the proposed HWTST is a faster and more efficient than HWBST, KS and T in terms of the computation time, error, accuracy, and distance of e.c.d.f.