Problem in Visual SLAM algorithm

HI,

I am using a stereo camera, gps, imu with laser scanner to find pose estimation on a moving vehicle. When i run the example code from my custom data obtained from a normal traffic road scene, some iterations are not completed due to error 'max number of trials reach, decreasing the desired confidence.' Also, i created bag of features for my dataset. I got follwoing problems

1. In some scenes, VSLAM got error as features are not found. so i discard the complete dataset and take new scene.
2. Due to moving cars and bikes near and far for stereo camera, comeplete travel distance, poses are not calculated for some images and algorithm terminated there.
3. The graph does not show complete trajectory as indicared with red line, however, the code is still running and left right images are compared.
4. Scale factor is 1.2, it is fixed for every test or how can i assume value for my test?
5. Parameters which need to be tweak for my data. in eaxmple numSkipFrames = 5 and numPointsKeyFrame = 80 are set.