Wireless target location technology has been widely used in civil and military fields. In the two-step localization algorithms, the signal measurements, such as thetime of arrival, theangle of arrival, the frequency difference of arrival, etc., should be extracted first from the receivedsource signal. Then the target position is identified by calculating the location equation. Compared with the two-step localization algorithms, the direct position determination (DPD) method, which need not estimate the signal parameters and calculate the position step by step, but obtainsthe source position from the received signals directly based on the maximum likelihood criterion, has been shown to have a goodestimation accuracy and robustness, especially under low signal-to-noise ratio (SNR) conditions. So it has been widely studied in recent years and has made remarkable achievements in academic research. However, the DPD algorithm of wideband signals emitters is not performing well with moving receivers in the joint positioning based on time delay and Doppler shift under the low SNRs. To obtaina better positioning performance, in this paper we present a DPD algorithm with variable velocity receivers based on coherent summation of short-time signal segments, and derive the source position Cramer-Rao lower bound (CRLB). The algorithm designs a positioning model in which the multiple variable velocity receivers are usedtoobtain the source signal, then the signal received at the same receiver is patitioned into multiple non-overlapping short-time signal segments, based on which, an approximate maximum likelihood estimator for the new DPD algorithm is developed. The algorithm makes full use of the location information contained in the coherency among the signals segments, while extra target position information is acquired through the speed variability in the positioning model, and thus the problem of location ambiguity is solved. The simulation results show thatthe algorithm proposed in this paper further improves the positioning performance, and outperforms the traditional DPD algorithms with more accurate results. Especially in the low SNR, it is closer to the CRLB.