Watershed Segmentation for Vehicle Classification and Counting

A robust video based system for the traffic surveillance system on the highway for vehicle detection, vehicle classification and counting for effective traffic analysis using only a single standard camera. The key goal of the proposed work is to successfully detect, track, classify and count the vehicle in partial occlusion and connected together by shadow on the highways. Marker-controlled watershed segmentation method is initially used for the extraction of the foreground regions from the highway scene. For tracking Gabor filter is applied which is used to measure vehicle path in video sequences. For effective vehicle classification support vector machine is utilized. An experiment result shows the considerable performance of watershed segmentation in vehicle detection in the occluded and connected together by shadow in the highways environment. Safety development of transport develops by computer vision. Nowadays it's widely used in several applications like transportation, military. Especially it is more useful in intelligent transportation. For the video based traffic analysis some of the parameters need to be performed they are vehicle detection, vehicle tracking, vehicle classification and counting .while performing such task there exist some issues like intensity changes, partial occlusion, missed vehicle due to darker region, connected together by shadow. Some of the existing techniques like magnetic loop detector, radar, laser, GPS are used for the intelligent transportation but the outputs are not much effective and also too high cost. Accuracy of the output also depends on the camera. Issues arise over camera are camera calibration, camera cost, and camera quality. Some of the approaches does not works fine in real time applications.(1) From the cameras the road lanes are need to be supervised to perform the process to analyze the traffic .For this, stationary camera is used in the purposed work. One single stationary camera is used and detect the vehicles provide the considerable result. From the camera the videos are captured and the frames are extracted into frames which are then converted to gray scale image and used for the further process (2). The important process of the vehicle detection is segmentation .Initial process of computer vision in intelligent transportation is background subtraction. Background subtraction has two approaches they are static and adaptive. In the static approach, dynamic changes cannot be updated but in adaptive approach issues like intensity changes, partial occlusion, missed vehicle due to darker region, connected together by shadow can be detected. Background subtraction approaches like Mixture of Gaussian, Frame Difference, and Appropriate Median results are compared with watershed transform segmentation algorithm results are shown in experiment results. For better result in segmentation another technique is sobel edge detection .In the binary images edges are found and connected together for efficient vehicle detection .It is used to detect the vehicles only in traffic scene because in traffic jam there may exists the pedestrians and also to avoid the misdetection of vehicles due to occlusion and shadow .There are two types of approaches in computer vision technology in intelligent transportation they are commercial approach ,sophisticated approach(2) the propose work merges these two approach with considerable cost and considerable output. Vehicle tracking is used to measure the current position of the moving vehicle in the path. And also it used to reduce noise which helps to improve the performance.Kalman filter, neural network, fuzzy measure also used for vehicle detection. The drawback of Kalman filter is misdetection of the vehicles in occlusion. In neural network it cannot find the global minimum. The changes in intensity cannot be adopted by fuzzy measure. Gabor filter is used in proposed work for effective vehicle tracking. Efficient background subtraction, vehicle tracking is used for the further process of vehicle classification and counting. For classification, support vector machine is used for the vehicle classification. In the proposed work the vehicles are classified as light vehicle and heavy vehicles. Light vehicles include the two wheelers and car. Heavy vehicles include truck, bus. When compared to other classification technique like SOM, MCMC, 3D classification and other manual classification techniques SVM classifies the vehicle effectively. For counting process the binary images is scanned. In proposed work two variables are used such as count, count reg. The process of vehicle detection, vehicle classification, vehicle counting is us ed to analysis the traffic and also it is