Safety is the most important for transportation, when we are driving, we constantly pay attention to our environment, as our safety and that of many other people are at stake. We particularly look out for position of potential obstacles, whether they are vehicles, pedestrians, or some events like work zone, constructions on the road. In this case, any advance in object detection system in vehicle can help lower the risk and power up the safety and help the autonomous vehicle to understand the surroundings.
There are many state-of-the-art technologies applied into image based autonomous vehicle detection, the very common technologies are the applications of deep learning models. The combination of HOG and SVM is good for vehicle detection when the traffic is very congested. YOLO is good for real time detection. Faster R-CNN is efficient and accurate in object detection. Mask R-CNN works great for pixel level object detection.
We are interested at understanding how deep neural network works for object detection since the mechanism as clear as conventional computer vision technique. Thus, we want to implement faster R-CNN with different core CNN structure and compare how it affect the efficiency and accuracy of models.