Lane Image Detection Based on Convolution Neural Network Multi-Task Learning

Based on deep neural network multi-task learning technology, lane image detection is studied to improve the application level of driverless assisted driving technology and reduce traffic accidents. The line database published by Caltech Tucson company used extract ROI (Region Interest), scale, inverse perspective transformation as well preprocess image, so enrich data set efficiency algorithm. In this study, ZFNet replace basic networks VPGNet, their structures are changed efficiency. Multi-label classification, grid box regression object mask three task modules build a named ZF-VPGNet. Considering that will be combined with embedded systems in future, compressed CZF-VPGNet without excessively affecting accuracy. Experimental results show vision system study achieved good test results. case fuzzy missing mark, improved algorithm can still detect obtain correct results, achieves high accuracy robustness. achieve real-time performance (26FPS), single forward pass takes about 36 ms or less.