The ubiquity of mobile devices with high-fidelity sensors and the sharp decreases in the cost of ultra-broadband wireless network flourish the market of spatial crowdsourcing, which has been proposed as a new framework to assign location-aware tasks (e.g., reporting road traffic, delivering food) to workers (i.e., persons equipped with smart device and willing to perform tasks). This paper studies the task assignment problem that concerns the optimal strategy of assigning each task to proper worker such that the total number of completed tasks can be maximized while all workers can go back to their starting point before expected deadlines after performing assigned tasks. It is an intractable problem since optimal assignment for individual worker does not necessarily lead to global optimal results. Observing that the task assignment dependency only exists amongst subsets of workers, this study utilizes tree-decomposition technique to separate workers into independent clusters and develops an efficient depth-first search algorithm with progressive bounds to prune non-promising assignments. Extended experiments demonstrate the effectiveness and efficiency of the proposed solution.