Many state of the art object classification applications require many data samples, whose collection is usually a very costly process. Performing initial model training with synthetic samples (from virtual reality tools) has been proposed as a possible solution, although the resulting classification models need to be adapted (fine-tuned) to real-world data afterwards. In this paper, we propose to use an incremental learning from cognitive robotics, which is is particularly suited for perceptual problems, for this bootstrapping process. We apply it to a pedestrian detection problem where a synthetic dataset is used for initial training, and two different real-world datasets for fine-tuning and evaluation. The proposed scheme greatly reduces the number of real-world samples required while maintaining high classification accuracy. We also demonstrate an innovative incremental learning schemes for object detection which training object and background samples one after the other: this keeps models simple by representing only those background samples that can actually be confused with pedestrians.