Research On Target Detection Of Pedestrian And Electric Motorcycle

With the continuous improvement of China’s economic level,China’s car ownership and electric vehicle ownership are also soaring,while cars and electric vehicles bring convenience to residents,traffic accidents are also growing.In view of the high mortality of pedestrians and electric motorcycles in traffic accidents in China,the research on real-time identification of pedestrians and electric motorcycles is of great practical significance for reducing mortality and improving travel safety.Through investigation and comparison,this paper puts forward the idea of using YOLOV3 network to identify electric motorcycles and to detect the position of electric motorcycles and pedestrians based on emergency braking area.The main research contents are as follows:Firstly,the convolution neural network based on CNN is analyzed,and YOLOV3 network with high recognition accuracy,good recognition for small targets and fast processing speed is selected according to the requirements of the recognition object in this paper.According to the research content of this paper: Pedestrian and electric motorcycle recognition research set the object of this data set collection.In order to enrich the data set,this paper chooses different places and different times to collect video,and then converts video into pictures.After screening the pictures,54211 pictures were collected.Finally,according to the training environment and recognition requirements,the hardware and software are configured.After 40,000 steps of training,the average loss of YOLOV3 network reaches 0.042549,and the training results meet the requirements.In this paper,the speed of 20 km/h and 40 km/h are selected as the experimental speed.According to the speed and braking distance,the dangerous area,warning area and safety area are set.Then,the pedestrians and electric motorcycles in 1080 P driving recorder video are discriminated,and the collision between the electric motorcycles and pedestrians is judged according to the area where the electric motorcycles and pedestrians are located.Through the statistics of the recognition rate,missed detection rate and false detection rate of pedestrians and electric motorcycles in video,the network trained in this paper can accurately and effectively identify pedestrians and electric motorcycles,with the lowest accuracy rate of95.7% and accurate position judgment;the processing speed is about 25 FPS,which is not much different from the real-time(30 FPS),meeting the real-time requirements,and achieving the purpose of this paper.