End to End Autonomous Driving Behavior Prediction Based on Deep Convolution Neural Network

The end-to-end automatic driving behavior prediction has become an important research direction in the field of automatic driving because of its simplicity and efficiency. Most of the existing end-to-end driving behavior prediction models use simple CNN structure. However, this method is vulnerable and captures less deep information, resulting in poor accuracy. In order to achieve more accurate end-to-end automatic driving behavior prediction, we combined the attention mechanism with the depth network and developed a residual network (ResNet50) model integrating the effective channel attention mechanism (ECANet). First, the residual network is used to extract spatial features from the RGB images collected by the left, middle and right cameras, and the effective channel attention module (ECA) is embedded to weight the attention of each feature channel. Secondly, the steering angle prediction result is output by using the weighted spatial feature information of the full connection layer fusion. Finally, an experiment was conducted using Udacity's public data set, which showed that the accuracy of ECA resnet50 in driving behavior prediction was better than other CNN models. In addition, compared with the model based on other attention mechanisms, its accuracy is also the highest.