Automobile Dashboard Object Recognition for Interested Vehicle in ITS

This paper proposes an algorithm to describe the distribution of automobile dashboard object (ADO) . In this algorithm, we propose ADO encoding and similarity computation algorithm SDEA. We encode every automobile dashboard object into dashboard cell to assign them specified anti-noise ability even if the ADO is interfered by environment light, dust, shadow and so on. Usually the ADO of every vehicle is different so that every vehicle has its own different presentation code. The encoding method present the ADO with accurate cells to decrease noises, and the difference between source and candidates can then be computed effectively. In this paper, we proposed a novel Synthesized Distance of Mahalanobis and Euclidean distance for Encoded ADO (SDEA) algorithm to minimize the correlation between cells of the ADO and increase the robustness of recognition. The tiny difference of the ADO is then enlarged and the recognition of an interested candidate is easier with strong robustness. In special vehicle detection, the algorithm has more effective performance compared with only Euclidian or Mahalanobis distance. Over 500 vehicle images are utilized in our experiments and the result verifies that the ADO cells encoding SDEA algorithm has distinguished recognition ability even if the images are contaminated by strong noise and the ADO has shape transformation and position shift so that this algorithm presents the ADO mathematically and recognizes a specified vehicle accurately in huge amount of automobile images and is effective for interested vehicle recognition in big data with an excellent noise tolerant capability. This algorithm overcomes the shortcoming of other category methods when they recognize a special vehicle with the serious transformed object structure and strong noise.