Fig. 1. Split algorithm of nodes

Fig. 2. Frame of object detection algorithm

Fig. 3. Diagrams of feature channels. (a) Original image; (b) gradient magnitude; (c) LUV color channels; (d) histogram of oriented gradients

Fig. 4. Diagrams of generated proposal windows. (a) Original image; (b) detection result; (c) proposal windows

Fig. 5. Detection results on SAA dataset. (a) Images under varied background; (b) images with noise; (c) multiple scale images; (d) images under different illumination conditions

Fig. 6. Precision-recall curves of different methods on SenseAndAvoid dataset

Fig. 7. Effect of number of trees on precision

Fig. 8. Effect of random sampling on precision