Links table
a summary
1 introduction
2 Background: Discovering 3D Multi -directional objects
3 primary experience
3.1 Preparation of the experiment
3.2 notes
3.3 Summary and Challenges
4 Panopticus Overview
5 Detecting 3D Multi -directional organisms
5.1 Form design
6 spatial implementation adaptation
6.1 Performance of performance
5.2 Form adaptation
6.2 Execution scheduling
7 implementation
8 evaluation
8.1 Test and Data Group
8.2 Preparation of the experiment
8.3 Performance
8.4 durability
8.5 component analysis
8.6 Public expenditures
9 related work
10 discussion and future work
11 conclusion and references
9 related work
9 Related Adaptive Detection Systems. Many computer vision systems designed for resources restaurant devices have achieved an effective use of resources by adapting to video content or computing budgets [14, 15, 19, 22, 50]. Parallel to these developments, many works have focused specifically on the methods of discovering adaptive organisms [3, 28, 49]Facilitating valuable applications such as Edge Video Analytics or the augmented reality of the mobile phone. For example, almost [49] Feel a multi -blush framework that turns between a detector and tracker based on awareness of the characteristics of video content and resource adherence. Remix [28] The video content brings air conditioning by dividing images and applying nerve networks selectively by different standards. However, the current adaptive methods limit the support of 3D disclosure of resource -saving resources. This type of detection requires the processing of each view of the camera using different three -dimensional awareness possibilities, such as the enhanced estimate of the 3D site of organisms and speed. To treat each way optimally, Panopticus predicts its expected performance based on the short -term future dynamics in spatial distribution, which is very important for mobile phone scenarios. Despite Remix [28] Performance estimation is used based on the distribution of objects, and it depends on the long -term historical distribution and does not explain the characteristics of different organisms in the 3D, which is not appropriate to prepare our system
Discover 3D objects on edge devices. With the rapid progress of the edge computing, there is an increased demand to employ the detection of 3D objects on the edge devices. Deepmix [18] The restrictions imposed on edge resources dealt with delegating the dual -dimensional detection tasks to a server equipped with high -performance graphics processing units. Other light light tasks are treated, such as 3D estimated location of the detected objects using a depth sensor, efficiently by mobile devices. Another solution, points [37]The parallel treatment technology that uses NPU and GPU Edge suggested to facilitate implementation on the device for 3D RGB-D. This approach shows the direction of harnessing the power of acceleration of artificial intelligence specialists to meet the requirements of edge computing [26, 27, 33, 59]. VIPs [44]It is designed for self -driving cars, providing an edge -based system that cooperates with the external infrastructure equipped with computing units and LIDAR sensors. This strategy has effectively extended vehicle visualization ranging by integrating data from the system on board and infrastructure. Aside from these efforts, Panopticus begins effectively in a comprehensive self -use system for resource restricted edge devices. The system has introduced a new concept of 3D Multi -directional perceptions, which led to the ability of computing on the edge, eliminating the need for depth sensors or emptying account.
This paper
