Research And Simulated Implementation Of Lateral Blind Spot Detection Based On Fisheye Camera

Because of economic development, the number of private cars grows rapidly in the past30years, but there are more traffic accidents followed. Some of the accidents happened because the drivers can not see the blind spot when they want to change lanes or turn their vehicles. If the drivers are warned when vehicles enter the blind spot, the accidents can be reduced. Now, developed countries begin to research driver assistance system to solve this problem. The video sensor is cheaper, more adaptable and has a wider scope compared to radar and infrared sensor. So researchers are high on the development and research of ADAS based on computer vision.This paper study detection technology of the blind spot based on fish-eye camera that used in the parking system. Through analysis of existing technology, paper presents lateral blind spot detection technology based on fisheye camera. The detection is divided into two stages:hypothesis generation and hypothesis verification. In the stage of hypothesis generation, preprocess the input image into gray image. And select the detection area to reduce the computation and improve the results. Then find the optical flow of the vehicles base on LK optical flow, select the feature points using frames subtraction, reverse tracking and cluster methods. That can find the location of the car quickly for the next stage. In the stage of hypothesis verification, verify the hypothesis from result of the first stage based on the shadow of the vehicle.Overall, this paper studies the technology and theory of the vehicle detection, and implemented the detection system. Experimental results show that the algorithm has high robustness and real-time work in the complex scenes.