Active cognition of affordance of household tool parts is regarded as an important aspect to improve home service robot intelligence. In order to meet the needs of real-time task of the service robot, a fast algorithm to improve the efficiency of affordance detection is proposed based on structured random forest (SRF). In the offline training phase, SRF is used to train affordance edge detector and affordance detector. Then the corresponding coarse-to-fine threshold of each affordance is determined by evaluating the results Fβof affordance detection. In the online detection phase, the affordance edge detector is used to calculate the initial probability map of the edge of affordance region. Then the coarse-to-fine threshold is used to obtain an outer rectangular including the region of the tool parts of corresponding affordance. Finally, the affordance detector is used to detect affordance of tool parts in the region obtained. The experimental results show that compared with the existing global search detection methods under normal non-grapnic processing unit systems, the average detection efficiency of the proposed method increases obviously, and the recall and precision are also improved.