《自动驾驶中的计算机视觉ppt课件.ppt》由会员分享，可在线阅读，更多相关《自动驾驶中的计算机视觉ppt课件.ppt（18页珍藏版）》请在三一办公上搜索。
1、1,Computer Vision for Autonomous Vehicles,2,Content,Object DetectionSemantic SegmentationReconstructionMotion & Pose EstimationTrackingScene Understanding,3,1. Object Detection,2D Object Detection 3D Object Detection from 2D Images 3D Object Detection from 3D Point Person Detection,Reliable detectio
2、n of objects is a crucial requirement to realize autonomous driving. The object detections include:,4,1. Object Detection,A paper list of object detection using deep learning from 2014 to now(2019),5,1. Object Detection,Disscusion,Object detection works already quite well in case of high resolution
3、with little occlusions. Remaining major problems across tasks are detection of small objects and highly occluded objects. Furthermore, a large amount of distant objects needs to be detected in some cases which is still a challenging task for modern methods since the amount of information provided by
4、 these objects is very low.,6,2. Semantic Segmentation,The goal of semantic segmentation is to assign each pixel in the image a label from a predefined set of categories. The task is illustrated in Figure 12 with all pixel of a certain category colorized in as specific color in a scene of the Citysc
5、apes dataset by Cordts et al. (2016) recorded in Zurich.,7,2. Semantic Segmentation,2.1 Semantic Instance Segmentation The goal of semantic instance segmentation is simultaneous detection, segmentation and classification of every individual object in an image.,The instance segmentation task is much
6、moredifficult than the semantic segmentation task. Each instance need to be carefully annotated separately whereas in semantic segmentation groups of one semantic class can be annotated together when they occur next to each other. In addition, the number of instance varies greatly between different
7、images.,8,2. Semantic Segmentation,2.2 Semantic Segmentation with Multiple Frames As autonomous systems are typically equipped with video cameras, temporal correlation between adjacent frames can be exploited to improve segmentation accuracy, efficiency and robustness.,9,2. Semantic Segmentation,2.3
8、 Semantic Segmentation of 3D Data 2D images lack important information such as the 3D shape and scale of objects which are strong cues for object class segmentation and facilitate thedetection and separation of individual object instances.,10,2. Semantic Segmentation,2.4 Semantic Segmentation of Str
9、eet Side Views One important application of semantic segmentation for autonomous vehicles is to segment street-side images (i.e., building facades) into its components (wall, door, window, vegetation, balcony, store, mailbox etc.). Such semantic segmentations are useful for accurate 3D reconstructio
10、n, memory-efficient 3D mapping, robust localization as well as path planning.,11,2. Semantic Segmentation,2.5 Semantic Segmentation of Aerial Images The aim of aerial image parsing is the automated extraction of urban objects from data acquired by airborne sensors. Aerial image parses can be used to
11、 automatically build road maps (even in remote areas) and keep them up-to-date. Furthermore, information from aerial images can be used for localization.,12,2. Semantic Segmentation,2.6 Road Segmentation,Segmentation of road scenes is a crucial problem in computer vision for applications such as aut
12、onomous driving and pedestrian detection. For instance, in order to navigate, an autonomous vehicle needs to determine the drivable free space ahead and determine its own position on the road with respect to the lane markings. However, the problem is challenging due to the presence of a variety of d
13、ifferently shaped objects such as cars and people, different road types and varying illumination and weatherconditions.,13,3. Reconstruction,14,4. Motion & Pose Estimation,2D Motion Estimation Optical Flow3D Motion Estimation Scene FlowEgo-Motion Estimation,15,5. Tracking,In tracking, the goal is to
14、 estimate the state of one or multiple objects over time given measurements of a sensor.Challenges: 1. Objects are partially or fully occluded by other objects or themselves. 2. The resemblance of different objects. 3. The interaction of objects in case of pedestrians further increases the amount of
15、 occlusions and makes it difficult to track each individual object. 4. Difficult lighting conditions and reflections in mirrors or. windows,16,6. Scene Understanding,One of the basic requirements of autonomous driving is to fully understand its surrounding area such as a complex traffic scene. The c
16、omplex task of outdoor scene understanding involves several sub-tasks such as depth estimation, scene categorization, object detection and tracking, event categorization, and more.,17,Challenge,How to improve generalization ability of the model. （如何提高模型的泛化能力）How do you make use of small-scale and super-large-scale data （如何利用小规模和超大规模数据）Comprehensive scene understanding （全面场景理解）Automatic network design （自动化网格设计）,18,谢谢!,
