Key Technology Research For Recognition Of Partially Occluded Objects

Recognition of partially occluded objects is the common problem in computer vision and pattern recognition and it is difficult to be solved. So it is very important to do in-depth research aiming at this problem. It is critical to choose a partial shape description method which is invariant in the rotation, translation and scale transform during the recognition of partially occluded objects.In this paper, we do some in-depth researches about the partial invariant characteristic of corner points. Mainly including the following:(1) An improved Harris corner detection algorithm is proposed. The classical Harris comer detection algorithm is analysed in detail at first.Because the redundance and false corner points are often produced in this classical algorithm, which have a bad influence on the shape description. So an improved algorithm is proposed.This algorithm adopt two restricted conditions to eliminate the redundance and false corner points.Experiment results show that the improved algorithm eliminates the redundance and false corners effectively and makes the corners location correctly.(2) In this paper a shape description method based on corners' partial invariant features is put forword after the analysis and comparison of some shape description methods. The new method can describe both the partial shape and the target's whole shape by uniting the partial shape. In this method we fist locate the corners correctly and get the sequence of corner by corner detection, secondly arrange the sequence and get the sequential corner sequece, thirdly choose three sequential corners to make feature triangle from the sequential sequence, finally calculate the corner features using feature triangle and complete the shape description.(3) Hash table is applied in the recognition of partially occluded objects, so that the recognition speed is raised. First we change the corner features to hash address using the hash function and construct the model base combining with hash table, secondly the number of each object is made statistic, thirdly the matching ratio is calculated, finally the occluded objects can be recognized from the occluded image through the matching ratio.The experiments show the method presented in the paper is a good way to recognize the occluded objects. The algorithm is simple and effective.