Using Hough Transform And Conditional Probability Model To Detect Multi-object

Computer vision algorithms for individual tasks such as object recognition, detection and segmentation have shown impressive results in the recent past. With the deepening of the computer vision theory, many methods have been integrated into the real world applications. Particularly, object detection and recognition plays an important role in the intelligent robot system, which needs a robust real-time vision system to detect the dynamic obstacle and understand the surrounding environment.The main target of this paper is to find a new framework for multi-object detection. The traditional methods based on hough transform does not have any probabilistic meaning, so a new probability model has been constructed to explain the relationship between the voting elements and the hypothesizes. This kind of model can get rid of the problems caused by some fragile post-processing methods, especially when handling the situation of the crowded objects.The model proposed in this paper is based on the theory of hough transform. For obtaining a hough transform function for this model, this paper discusses on how to build a hough forest based on random forest. Besides, a review on different feature extraction methods such as scale-invariant feature transform and histogram of oriented gradients is given in the paper. And some popular random fields have been discussed, including conditional random fields and markov random fields.Finally, a greedy method has been proposed for the maximum a posteriori. The performance of the experiments demonstrates the correction of the method.