Vehicle Warning Algorithm Based On Fiy's Visual Mechanism And Its Application

Vehicle video sequence analysis in complex road scenes is an important issue in the field of intelligent traffic.It is an important topic in the field of traffic safety how to process video images in real time,quickly and effectively,which involves how to make full use of real-time information of moving objects in images to accurately and effectively send safety alarm information.Fruit fly is an insect who is sensitive to moving objects.Its visual system can detect incoming collision objects in real time and accurately,and can effectively accomplish tasks such as objects recognition,objects tracking,collision detection and avoidance.Inspired by this and based on the studies of image stabilization,weather recognition,target extraction and target tracking,this thesis discusses a vehicle safety collision warning algorithm based on fruit fly's visual biology principle,while transplanting and testing the algorithm on DSP hardware platform.The main works and achievements acquired are summarized as follows:1.For the problem of image stabilization,an improved gray projection image stabilization algorithm is developed based on the idea of uniform image division and gray projection.On the basis of a reported weather recognition algorithm and an existing target extraction algorithm,an improved target extraction algorithm is proposed to solve the problem of target extraction,in terms of three kinds of image features,i.e.standard variance,smoothness and image entropy.Subsequently,an improved collision warning algorithm is designed by fusing such two algorithms into a reported fly visual neural network collision warning model.With video images originated from actual scenes,comparative experiments show that the collision warning algorithm can obtain better warning effect.2.For the problem of vehicle collision warning in complex road scenes,two algorithms of targets' calibration and tracking are designed to determine target vehicles' number and their positions presented in a given image sequence;the variable characteristic of each target vehicle's position between two adjacent frames and the reported artificial fly visual neural network are adopted to calculate the slope of the moving direction of the target.Such two algorithms,together with a new collision warning scheme form a target tracking and collision warning algorithm.Comparative experiments show that the algorithm can effectively transmit warning signals in a timely and effective manner.3.The above target tracking and collision warning algorithm is transplanted into the DSP hardware environment,in order to obtain a vehicle collision warning prototype.Based on multiple kinds of video image sequences in different scenes,comparative experiments validate that the prototype can not only solve the problem of vehicle collision warning in different scenarios,but also send warning signals efficiently and effectively.