computer vision - Get good approximated segmentation mask within the detection rectangle -

i want fast , approximated segmentation of object within detection box. how that? tried background subtraction algorithms , grab cut dont promising.

@shai problem grabcut color not discriminative feature therefore times might give better segmentation other times. im using kinect sensor, able give depth info. im using deformable part model approach , have in end following input:

1. position , size of each deformable part
2. position of final detection box
3. rgb
4. depth

im providing sample rgbd image numbers:

final rect: - 128 x 352 (256, 64)

parts:

• 96 x 96 (304, 64)
• 96 x 96 (304, 128)
• 96 x 96 (320, 320)
• 96 x 96 (256, 224)
• 96 x 96 (272, 112)
• 96 x 96 (272, 320)
• 96 x 96 (272, 48)
• 96 x 96 (288, 224)