In order to identify and locate vehicle more quickly and accurately

an improved algorithm based on convolutional neural networks（ CNN） is proposed. The algorithm is optimized in the proposed network and feature extraction. Specifically

the framework is embedded with a light-weight proposal network to generate candidate region

and early discard candidate frame of impossible regions to reduce background numbers; feature fusion technology is used to extract hyper feature

refine the identification and location of the vehicle

which improve the quality and precision of vehicle detection effectively. Further verify on large-scale datasets of vehicle detection and tracking and the self-collected vehicle datasets

and compared with current advanced algorithms

compared with Faster RCNN

average detecting quality is improved by 9. 91 %

which fully shows the effectiveness of the algorithm.