Conclusion

Discussion

From our evaluation result, it is apparently that even we are only using simplified version of YOLO, we have decent prediction almost the same as the complete YOLO, while our model only take less than a quarter of complete one's runtime. However, one defect of our model is that it does not predict pedestrians and trucks as well as cars owing to insufficient amount for training data in pedestrians and trucks.

Future Work

Undoubtedly we will train our model with more classes and more object instances per class to increase the number of different objects able to be predicted and improve the recall of each object being detected. Also, combining with other algorithms and models to detect passable lanes and to follow the traffic rule by understanding the road signs and traffic lights is provisional to improve the quality of self-driving car.