We present a fast and simple approach to pedestrian detection and tracking based on shape features. The task is performed in two steps. For pedestrian detection, we use the difference of temperature between the background and target pedestrians in thermal images. The proposed temperature-based threshold method can detect the precise location and shape of pedestrians. In the tracking step, we extract the Histogram of Oriented Gradient (HOG) as a local shape feature. The transition score between the adjacent frame's detected pedestrians is calculated using a feature distance measure. Smaller feature distances lead to higher transition scores. Performance is evaluated on the public thermal image benchmark dataset OTCBVS. The proposed algorithm performs multi-pedestrian detection and tracking effectively. In this paper, we leave occlusion and long-term tracking issues for further work.