During the race, the football, goal and two sides of the robot need to be recognized by the vision system of humanoid robot soccer. Considering the rapidity and effectiveness of the algorithm, the ball and the goal are identified by the algorithm based on color information. The possible positions of the ball and the goal are extracted by the color thresholds of the ball and the goal in the picture, then the correct position is determined by the background color or area information. To recognize the two sides of the robot, first the features of the robot should be extracted and then through online real-time supervision learning method a cascade classifier is trained, which is used to identify robots. Result shows that the field objects can be recognized quickly and effectively and the algorithm has better robustness.