Application Of Mathematical Morphology And Wavelet Analysis In Intelligent Vehicle Detection Systems

The common information platform for Intelligent Transportation Systems (ITS) is the central nerve of the modern transport system, which plays a monitoring, scheduling, command and planning functions in the entire transport system. Traffic information collection technology is the basis to constructing the ITS common information platform, in which the most critical and complex part is the automatic detection technology to vehicles. Automatic visual inspection technology using machine vision instead of artificial vision for target extraction and recognition has the advantage of both machines' continuous work and human cognitive ability. Thus, vehicle detection technology based on automatic vision is becoming the mainstream of the field study.In the past few years, the laboratory conducted a lot of research work and constructed a complete set of vehicle Detection Algorithm based on images' background Difference. Lots of practice showed that the algorithm worked well in simple traffic scene, but in the complex environment its efficiency and accuracy were yet to be enhanced. Aimed at the detection problems under the complex environment, in the algorithm some improvement strategies were made, but which mainly focused on the physical characteristics of the image targets. However, image geometry characteristics are nearly not impacted by the outside environment. If combining the physical and geometry characteristics to analysis image will be bound to improve the accuracy and adaptability of the algorithm. In this paper, the difficulties and several issues to be resolved in the practical application are analyzed in detail, and the theoretical research direction based on mathematical morphology and wavelet analysis is proposed. It focuses on studying the key algorithms of noise filtering, connected object segmentation, vehicles tracking by target matching. The experimental results show that the algorithms proposed in this paper are practical and effective.Aiming at the image noise problems caused by poor background quality, light change in the environment, shadow elimination and jitter of video equipment, the CS morphology filtering algorithm based on extension measurement of target is proposed. In the algorithm, the extension concept is used to distinguish the noise signal and detail signal effectively for their morphological difference, which preserves the filtering from damaging image details. And at the same time, a morphological hit operator based on contour-structuring element is defined. It can achieve the Synchronized elimination to positive and negative noise. It is used to handle the connectivity components of image, which can realize the regional filtering. Finally, in order to effectively deal the cases with intensive noises or big noise points, the iterative filtering strategy with multi-scale structuring elements is proposed. CS morphological filter weakened the dependence of morphological operators to the form of structuring elements. As a conclusion, CS morphological filter weakened the dependence of morphological operators to the form of structuring elements, and the regional filtering method makes the operational efficiency increased greatly. The experimental results show that CS filter can suppress image noise effectively, at the same time retain the details from loss.Aiming at the problem that CS morphological filtering effect would be impacted by the noise size and its surrounding, the improved combination solution of morphological wavelet and CS morphological filter is proposed. Morphological wavelet has the characteristic of multi-resolution decomposition to images, and can decompose the energy and form of image noise signals to multiple directions, which makes the noise signals meet the conditions of CS filtering easily. Therefore, the CS filtering result to image decomposed by morphological wavelet can be comparable to the results by the iterative filtering strategy with multi-scale structuring elements. The experimental results show that the combination solution of the two can improve the filtering effect and has a stronger applicability.Aiming at the problem that the differential algorithm would undermine the integrity of vehicle target, and combining the feature that vehicle has convex-border form, a convex reconstruction algorithm with two-level hits is proposed. Under the premise with no change to the convex or concave shape of the objectives, to binary image, the algorithm can fill up the hollow inner the objective and its concave with multiple directions on the edge; to gray-scale image, it can merged the darker regions and texture of the objective into the bright surrounding regions gradually.Aiming at the overlapping problem of vehicles caused by projected shadow or objectives' occlusion, an overlapping segmentation algorithm based on morphological watershed is studied. By the reasonable integration of convex reconstruction, distance transform and watershed algorithms, a valid method considered to not only gray shading characteristics but also morphological form distribution is proposed. It uses the convex reconstruction and the gray-scale grading operations to smooth the image, which make its gray-scale more uniform and easier to be segmented. On the other hand, do threshold operation on the smooth gray-scale image, and then extract the objectives' own unique morphological identifier by using distance transform. At this time, take the smooth gray-scale image and the objectives' morphological identifier image as the input parameters of watershed algorithm to complete the overlapping segmentation. The experimental results show that as long as the overlapping objectives have enough gray-scale difference or some connection curvature on their form, they can be segmented.The traditional image matching method by measuring the accumulation of gray-scale difference always has little robustness for the causes such as scene occlusion and too strong or too weak light. Aiming at this problem, an image matching algorithm by measuring morphological size is proposed. It inspects the matching degree by measuring the morphological size of gray-scale difference set between the objectives. This method considers not only the gray-scale difference between but also the geometric feature of the difference set, so its anti-noise performance has been enhanced greatly comparing to the classic matching algorithm. During the course of computing morphological size, threshold decomposition strategy of gray-scale function is adopted, which transformed gray-scale morphological operations to binary morphological operations. This makes the realization of the algorithm easier and less computation. Also, quick interval approximation strategy is adopted to accelerate computing speed. The experimental results show that the morphological matching algorithm can accelerate the matching process and has a high reliability and stability.