Intelligent Driving Perception Interval Labelling Algorithm

AI is the soul of intelligent driving, intelligent driving AI technology needs a lot of annotation data, its high cost limits the development of intelligent driving, and most of the data is discarded because of the high annotation cost. In this paper, an intelligent driving perception interval labeling algorithm is proposed to obtain scene complete data labeling at low cost. The algorithm deduces other frames based on the prior information of a small number of interval labeling frames, and deduces truth frame by using multi-frame point cloud, interpolation frame and detection frame. Combining the advantages of interpolation frame and rough frame, it can maintain high precision and high recall rate under difficult conditions, and greatly reducing the cost of labeling. The experiment shows that the recall rate of IOU 0.7 with an interval of 10 frames is as high as 0.986, which increases by 31 and 15 percentage points respectively on the basis of the detection frame and the interpolation frame, and there is basically no missing and false detection phenomenon, and the detection results hardly need manual adjustment. Even if the interval of 30 frames is marked with one frame, the IOU 0.7 recall rate reaches 0.87, which has great reference value for reducing the marking cost of intelligent driving.