Abstract

Driver assistance systems in modern cars now show clear steps towards autonomous driving and improvements are presented in a steady pace. The total number of sensors has also decreased from the vehicles of the initial DARPA challenge, more resembling a pile of sensors with a car underneath. Still, anyone driving a tele-operated toy using a video link is a demonstration that a single camera provides enough information about the surronding world.   Most lane assist systems are developed for highway use and depend on visible lane markers. However, lane markers may not be visible due to snow or wear, and there are roads without lane markers. With a slightly different approach, autonomous road following can be obtained on almost any kind of road. Using realtime online machine learning, a human driver can demonstrate driving on a road type unknown to the system and after some training, the system can seamlessly take over. The demonstrator system presented in this work has shown capability of learning to follow different types of roads as well as learning to follow a person. The system is based solely on vision, mapping camera images directly to control signals.   Such systems need the ability to handle multiple-hypothesis outputs as there may be several plausible options in similar situations. If there is an obstacle in the middle of the road, the obstacle can be avoided by going on either side. However the average action, going straight ahead, is not a viable option. Similarly, at an intersection, the system should follow one road, not the average of all roads.   To this end, an online machine learning framework is presented where inputs and outputs are represented using the channel representation. The learning system is structurally simple and computationally light, based on neuropsychological ideas presented by Donald Hebb over 60 years ago. Nonetheless the system has shown a cabability to learn advanced tasks. Furthermore, the structure of the system permits a statistical interpretation where a non-parametric representation of the joint distribution of input and output is generated. Prediction generates the conditional distribution of the output, given the input.   The statistical interpretation motivates the introduction of priors. In cases with multiple options, such as at intersections, a prior can select one mode in the multimodal distribution of possible actions. In addition to the ability to learn from demonstration, a possibility for immediate reinforcement feedback is presented. This allows for a system where the teacher can choose the most appropriate way of training the system, at any time and at her own discretion.   The theoretical contributions include a deeper analysis of the channel representation. A geometrical analysis illustrates the cause of decoding bias commonly present in neurologically inspired representations, and measures to counteract it. Confidence values are analyzed and interpreted as evidence and coherence. Further, the use of the truncated cosine basis function is motivated.   Finally, a selection of applications is presented, such as autonomous road following by online learning and head pose estimation. A method founded on the same basic principles is used for visual tracking, where the probabilistic representation of target pixel values allows for changes in target appearance.

Abstract

Vision based active safety systems have become more frequently occurring in modern vehicles to estimate depth of the objects ahead and for autonomous driving (AD) and advanced driver-assistance systems (ADAS). In this thesis a lightweight deep convolutional neural network performing real-time depth estimation on single monocular images is implemented and evaluated. Many of the vision based automatic brake systems in modern vehicles only detect pre-trained object types such as pedestrians and vehicles. These systems fail to detect general objects such as road debris and roadside obstacles. In stereo vision systems the problem is resolved by calculating a disparity image from the stereo image pair to extract depth information. The distance to an object can also be determined using radar and LiDAR systems. By using this depth information the system performs necessary actions to avoid collisions with objects that are determined to be too close. However, these systems are also more expensive than a regular mono camera system and are therefore not very common in the average consumer car. By implementing robust depth estimation in mono vision systems the benefits from active safety systems could be utilized by a larger segment of the vehicle fleet. This could drastically reduce human error related traffic accidents and possibly save many lives. The network architecture evaluated in this thesis is more lightweight than other CNN architectures previously used for monocular depth estimation. The proposed architecture is therefore preferable to use on computationally lightweight systems. The network solves a supervised regression problem during the training procedure in order to produce a pixel-wise depth estimation map. The network was trained using a sparse ground truth image with spatially incoherent and discontinuous data and output a dense spatially coherent and continuous depth map prediction. The spatially incoherent ground truth posed a problem of discontinuity that was addressed by a masked loss function with regularization. The network was able to predict a dense depth estimation on the KITTI dataset with close to state-of-the-art performance.