Autonomous Driving Challenge: To Infer the Property of a Dynamic Object Based on Its Motion Pattern

In autonomous driving applications a critical challenge is to identify the action to take to avoid an obstacle on a collision course. For example, when a heavy object is suddenly encountered it is critical to stop the vehicle or change the lane even if it

  • PDF / 1,189,436 Bytes
  • 7 Pages / 439.37 x 666.142 pts Page_size
  • 73 Downloads / 231 Views

DOWNLOAD

REPORT


Abstract. In autonomous driving applications a critical challenge is to identify the action to take to avoid an obstacle on a collision course. For example, when a heavy object is suddenly encountered it is critical to stop the vehicle or change the lane even if it causes other traffic disruptions. However, there are situations when it is preferable to collide with the object rather than take an action that would result in a much more serious accident than collision with the object. For example, a heavy object which falls from a truck should be avoided whereas a bouncing ball or a soft target such as a foam box need not be. We present a novel method to discriminate between the motion characteristics of these types of objects based on their physical properties such as bounciness, elasticity, etc. In this preliminary work, we use recurrent neural network with LSTM (Long Short Term Memory) cells to train a classifier to classify objects based on their motion trajectories. We test the algorithm on synthetic data, and, as a proof of concept, demonstrate its effectiveness on a limited set of real-world data.

1

Introduction

In recent years, the technology of self-driving cars has made dramatic progress. One of the critical challenges of this emerging technology is the safety of both car occupants and other road users. The current prototype of autonomous cars are equipped with advanced sensors such as ultrasonic, vision, radar and LIDAR. These sensors along with sophisticated data fusion algorithms are able to detect and track obstacles in real-time with very good resolution. When an obstacle is detected in the planned path, either its planned route should be modified or the vehicle should come to stop. Depending on the traffic situation and vehicle speed, this policy could cause collision with other vehicles. Therefore, obstacle avoidance may not always be the safest action. Similar challenge has been discussed in [7]. The intuitive solution would be to recognize the object before taking an action. The intelligent unit should predict whether it is safe to pass over the object or it should inevitably follow avoiding policy. c Springer International Publishing Switzerland 2016  G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part III, LNCS 9915, pp. 40–46, 2016. DOI: 10.1007/978-3-319-49409-8 6

Autonomous Driving Challenge: To Infer the Property

41

A sample video for each scenario is downloaded from Youtube and a few frames are shown in the Fig. 1. In the first video, an empty plastic container is bouncing in the road which is safe to pass. In the second video, a heavy object is falling out of the front car which should definitely be avoided.

(a)

(b)

Fig. 1. Selected frames of dynamic objects on the road. (a) A plastic container which is safe to collide, YouTube Link. (b) A heavy object that should be avoided, YouTube Link.

The immediate solution that one might consider is to formulate the problem as a regular image classification task and collect a dataset of collision safe and unsafe objects. While there is much prog