Research On Forward Merging And Collision Warning Algorithm Based On Image Sensor

The forward collision warning system is an important branch in the field of safety assisted driving.With the development of the automobile industry and the improve-ment of people's living standards,more and more private cars have entered people's lives.Some drivers have insufficient traffic safety awareness,They are distracted dur-ing driving and are free to change lanes,which may lead to traffic accidents.Research on forward merging and collision warning algorithm based on image sensor proposed in this paper uses the vehicle camera to obtain the image data of a certain area in front of the current vehicle,calculates and obtains the target vehicle through the obj ect detection and tracking method,the distance,speed,trajectory and category of the target vehicle is obtained by the vehicle camera calibration ranging,and comprehensively judging the information,thereby the front vehicle collision time warning,the preceding vehicle merging line warning and the non-motor vehicle warning are realized,when judging the potential danger,promptly reminds the driver to take effective measures through sound,touch or auxiliary control.The main work and results of this paper are as follows:1.A method of object detection in traffic scene based on deep learning is pro-posed.The object detection is the core of the entire collision warning algorithm,which determines the accuracy of the early warning system.The method of this paper uses deep neural network to extract target features,combined with the geometric features of the target(such as pedestrians,vehicles,etc.)in the traffic scene,uses rectangular anchors of appropriate size and number to detect the target in the image on two scales.Experiments show that This method effectively reduces the training and detecting time of the target detection algorithm while ensuring accuracy.2.An target tracking method of adaptive scale based on road texture context in-formation is proposed.The target tracking algorithm implements the functions of po-sitioning,trajectory recording and motion trend judgment of the target vehicle in front,According to the relative position of the target vehicle and the road texture area under the image,the stability of the road texture,and the noise areas above,left and right of the target,using the machine learning method,the tracking algorithm will feature the adjacent areas around the target according to certain weights,the online training and detection of the front target are realized.On this basis,the multi-scale tracking method is used to adjust the target's border according to the actual scale change.Ex-periments show that this method can effectively improve the positioning accuracy and anti-interference ability of the tracking algorithm.3.A fusion strategy of object detection and object tracking methods,a preceding vehicle merging line warning,collision warning and non-motorized warning strategy are proposed,this algorithm uses the queue data structure to save and update the ob-ject trajectory information,and realizes the front-vehicle parallel line warning function according to the traj ectory information;The front vehicle collision time warning is real-ized according to the collision time;According to the category of the object,the object is divided into motor vehicles and non-motor vehicles to realize the non-motor vehicle warning function.Experiments show that this strategy has high robustness and early warning rate.Especially in the case of elevated or high-speed roads,the front-vehicle parallel line warning system can detect the intention of the line and timely and timely warning.