Motion Boundaries Formation For Autonomous Vehicles

<div>In this paper, the application of Model Predictive Control to perform curvilinear motion planning is explored. More specifically, nonlinear MPC will be focused on because of its proven efficiency in the modeling of uncertainties as well as in nonlinear model dynamics. The main objective of this report is to show that with proper modeling and formulation of motion constraints, curvilinear motion planning can be achieved with nonlinear MPC. The trajectory of the vehicle will be tracked with the least error while satisfying constraints such as speed and steering angles. Simulations are presented which demonstrate the ability of the suggested models to successfully perform curvilinear motion staying safely within the bounds, while simulations of several models validate its performance. A deterministic sensitivity analysis was conducted in order to determine the impact</div><div>of the prediction horizon time. Experimental results show that a critical prediction horizon time approximately 10 to 13 seconds was identified as the ideal range for optimal results of the model.</div>

Two-stream fusion edge detection network

This thesis introduces a method to combine static and dynamic features in a convolutional neural network (CNN) to produce a motion and object boundary prediction map. This approach provides the CNN with dynamic and static cues and information, thus improving its predictions. The spatial stream of the CNN learns to compute an object boundary prediction map from a single RGB frame, while the temporal stream learns to compute a motion boundary prediction map from the corresponding optical ow map. The streams are then combined through an encoder-decoder architecture, where the decoder learns to fuse the features from both streams to obtain a task specific output. The proposed method yields state-of-the-art results on a motion boundaries benchmark, and systematic improvements in object boundaries benchmarks over methods that solely rely on static features extracted from a single RGB frame.

Two-stream fusion edge detection network

This thesis introduces a method to combine static and dynamic features in a convolutional neural network (CNN) to produce a motion and object boundary prediction map. This approach provides the CNN with dynamic and static cues and information, thus improving its predictions. The spatial stream of the CNN learns to compute an object boundary prediction map from a single RGB frame, while the temporal stream learns to compute a motion boundary prediction map from the corresponding optical ow map. The streams are then combined through an encoder-decoder architecture, where the decoder learns to fuse the features from both streams to obtain a task specific output. The proposed method yields state-of-the-art results on a motion boundaries benchmark, and systematic improvements in object boundaries benchmarks over methods that solely rely on static features extracted from a single RGB frame.

Processing of motion boundary orientation in macaque V2

Human and nonhuman primates are good at identifying an object based on its motion, a task that is believed to be carried out by the ventral visual pathway. However, the neural mechanisms underlying such ability remains unclear. We trained macaque monkeys to do orientation discrimination for motion boundaries (MBs) and recorded neuronal response in area V2 with microelectrode arrays. We found 10.9% of V2 neurons exhibited robust orientation selectivity to MBs, and their responses correlated with monkeys’ orientation-discrimination performances. Furthermore, the responses of V2 direction-selective neurons recorded at the same time showed correlated activity with MB neurons for particular MB stimuli, suggesting that these motion-sensitive neurons made specific functional contributions to MB discrimination tasks. Our findings support the view that V2 plays a critical role in MB analysis and may achieve this through a neural circuit within area V2.

Unsupervised Video Object Foreground Segmentation and Co-Localization by Combining Motion Boundaries and Actual Frame Edges

In this article the authors proposed a fast and fully unsupervised approach for a foreground object co-localization and segmentation of unconstrained videos. This article first computes both the actual edges and motion boundaries of the video frames, and then aligns them by the proposed HOG affinity map approach. Then, by filling the occlusions generated by the aligned edges, the paper obtained more precise masks about the foreground object. With an accumulation process, these masks could be derived as the motion-based likelihood, which is used as a unary term in the proposed graph model. Another unary term is called color-based likelihood, which is computed by the color distribution of foreground and background. Experiment results shows the method is fast and effective to detect and segment foreground objects.