The invention discloses an unmanned pedestrian trajectory prediction method based on a convolutional neural network. The method comprises the steps of processing sample data, obtaining an input information sequence, constructing and optimizing the network, and testing and evaluating an optimal model. Real-time video collected by a visual sensor on an unmanned vehicle is segmented into images withframes as units to serve as sample data, target crowds about to pass through zebra stripes in the sample data are divided into three classes, and pedestrian position-proportion information sequence, the pedestrian skeleton information sequence and the motion sequence of the visual sensor,are obtained from samples;, ; the information sequence is input into a convolutional neural network for training to obtain a preliminary prediction model, and after testing and evaluation, finally a prediction track and an action category are output. According to the method, the convolutional neural network isadopted to predict the trajectory of the unmanned pedestrian, so that the probability of pedestrian collision in the road driving process of the unmanned vehicle can be effectively reduced.