The series problem of infrared small target detection in heavy clutter is a challenging work in active vision. During different imaging environments the size and gray intensity of target will keep changing which lead to unstable detection. Focus on mining more robust feature of small targets and following the sequential detection framework, we propose a novel research scheme based on density-based clustering and backtracking strategy in this paper. First, point of interest is extracted by the speeded up robust feature (SURF) detector for its better performance in digging features invariant to uniform scaling, orientation and illumination changes. Second, due to the local aggregation property of target trajectory in space, a new proposed density-based clustering method is introduced to segment the target trajectory, so that the target detection problem is transformed into the extract the target trajectory. Then, In order to keep the integral and independence of the trace as much as possible, two factors: percent and are exploited to help deciding the clustering granularity. Later, the backtracking strategy is adopted to search for the target trajectory with pruning function on the basis of the consistence and continuity of the short-time target trajectory in temporal-spatial. Extended experiments show the validity of our method. Compared with the data association methods executed on the huge candidate trajectory space, the time-consuming is reduced obviously. Additional, the feature detection is more stable for the use of SURF and the false alarm suppression rate is superior to most baseline and state-of-arts methods.