extracting boundary of objects in real images

I work on real images taken by a robot, and i want to extract the boundaries of all objects in each image captured by the robot. I follow that sequence:

1. read image
2. rgb2gray
3. find suitable threshold
4. gray2bw
5. morphological operations
6. bwboundaries

*the problem is: during the processing most of the objects disappear and hence the output boundaries are for one or two objects only.

please help me find a robust way to extarct the boundaries from such real cluttered images. *

[Sean de Wolski]

And how do you expect us to do that without seeing your images?