Nighttime Front Vehicle Detection For Heavy Trucks

With the rapid development of economy, the vehicles have increased by a large margin, which facilitates our daily life greatly, while the possibility of traffic accidents also increases. It has becomes important research content of vehicle and automation how to improve safety of vehicles, and the Auto-auxiliary Driving System grows up with it. The Auto-auxiliary Driving System realizes the communication between vehicle and human or road and so on with advanced sensors installed on the car such as radar and camera to make the vehicle perceive the environment around, analyze the running state of this vehicle, run to the destination according to humans’will, and finally realize the automatic driving in the future.In day time, it is abundantly bright. The features of front vehicles are obvious, such as the color, the shape and the veins which makes it easy to detect a vehicle. But at nighttime, because of lack of bright, driver can’t locate the vehicle quickly and accurately. It has become a hot research field to realize intelligently detect vechicles at nighttime. Compared with the traditional night vision technique such as the infrared ray, radar and so on, the Auto-auxiliary Driving System based on computer vision has advances such as simple hardware equipment, rapid processing, low cost and so on, which is more suitable to common vehicles. Heavy trucks have features of high speed, big bulk and large inertia. It often makes a threat to other vehicles and pedestrian. Especially driving on highway at nighttime, it is difficult for heavy truck driver to locate other vehicles in short time, which makes heavy trucks more dangerous. It is very important for road safety to make heavy trucks detect front vehicles intelligently at nighttime. This paper study the front vehicles detection problem at nighttime for heavy truck in environment of vehicle testing field and highway.At nighttime, a pair of bright taillights is the most obvious feature of front vehicle, the symmetry of which is often used to detect vehicles. Under the influence of highlight of headlights of heavy truck, the taillights of front vehicle often don’t show absolutelysymmetry. The binocular vision, often use to realize front distance measurement, and often encounter problems of synchronization drift. So we shoot grey-level images of heavey trucks driving environment. The headlights of heavy truck are very bright, and it tends to make interference information on the road, road barrier or other buildings. Particularly for heavy trucks’ driving environment, this paper proposes an improved threshold algorithm, which can remove other interference information except the taillights as much as possible. This paper proposes an new vehicle detection method of combination of the improved threshold method and training a classifier:Cut training samples for training classifier from the threshold processed images, train classifier based on Haar-like and Adaboost algorithm, during detection process the current frame with the improved threshold method, finally detect vehicle on the threshold processed image and mark the detection result on the original image.To realize the improved detection method, this paper firstly represents Otsu, the histogram threshold method, the maximum entropy algorithm and the Kumar threshold algorithm. By analyzing and comparing these algorithms, choose a algorithm most suitable to the heavy truck driving environment and improve it to remove the interference information. This paper also describes the meaning of Haar-like, Haar-like templates and its fast calculation method. Then this paper show the Adaboost algorithm, including the boosting algorithm background, how to find weak classifier, the training flow of Adaboost, and the cascade method. Design experiment to verify the effectiveness of the improved threshold method and the improved vehicle detection method. We shoot images of vehicle testing field and highway, compare and analyze the processing results of the proposed threshold algorithm and other methods for our these images. Make training samples with images from testing field, training two classifiers by the traditional method and the improved method separately. Make two testing sample sets separately from vehicle testing field and highway, and analyze the detection performance of two classifier on the two testing sample sets. Add samples from highway to the training sample set, training two classifier separately by the traditional and the improved method. Testing the two new classifier on the two kinds of testing samples, and analyze all of the testing results. Finally, compare the detection performance of the improved detection method and the method based on the symmetry. The experiment results prove that the improved threshold algorithm is more suitable for our study environment and the improved training and detecting method is more accurate and more robust, and it is better than the method based on the symmetry, which makes great significance in improving road safety.