Driver Intention Recognition Based on Computer Vision

In order to improve the accuracy of driver intention recognition and recognize the driver's lane changing intention earlier, a driver intention recognition system based on computer vision is established in this paper. The system judges the driver's head posture based on the fine-grained structure aggregation network (FSA-Net), determines the driver's observation behavior of the rear-view mirrors on both sides during driving by comparing with the non lane changing stage, divides the identification section in combination with the vehicle parameters, and uses the conditional random field (CRF) identify and predict the driver's lane change intention, provide the driver's behavior data for controlling lane change and other behaviors in the driver assisted driving system, and improve driving safety. Through experimental comparison, the system can detect the driver's lane change intention 4.3 seconds before lane change; the overall recognition rate is more than 98%; the recognition time is 0.013026s, which meets the requirements of online recognition.