Vehicle Recognition Based On Optical Flow Algorithm

The vehicle recognition algorithm based on optical flow makes use of the continuous frames to estimate the three-dimensional motion state of the vehicle. This method has great importance in the driver assistant system by detecting and tracking the object to acquire the real-time environmental information from the front.As the camera and target both move, with vehicle body vibration, various illuminations, the object’s size change, etc, the front image collected from the moving vehicle can be considered as complex scene. This paper focuses on how to get the reliable optical flow and the analysis on the vehicle moving state according to the local characters and driving environment.The contributions of this dissertation are as follows.1. The comparison on the accuracy and computation speed between the algorithm based on Horn-Schunck and Lucas-Kanade are made to get the analysis of suitability on the usage scenario.2. As to the SIFT, the accuracy can be elevated to sub pixel. The optical flow, matched from feature points, may be adapted to the complex scene and visualized in the color space.3. The vehicle recognition method in complex scene is proposed in this paper. Focusing on the set of interested feature regions to generate SIFT optical flow field. Meanwhile, its fusions of vehicle shadow, lane position, FOE, driveway region and environmental characteristics have enhanced the robustness of the recognition and saved computing time, and finally the target vehicle motion state is judged(including vehicle position, relative driving speed and direction, vehicle distance).Experimental results from matlab illustrate the effectiveness of this method.Finally, the conclusion and future work are illustrated in the end.