Deep Learning for analyzing traffic camera video

点击翻译按钮获取中文摘要

The project involved improving the performance and enhancing the capabilities of the BriskVANTAGE trafficanalysis software from the BriskSynergies company. Modern state-of-the-art machine learning technology(deep convolutional neural networks) will be used to improve the detection and recognition of objects such asmotorcycles, bicycles, trucks and buses, in addition to automobiles and pedestrians. The deep net algorithmswill be modified to include robust object tracking and data association (retaining the identity of which sceneobject is which when tracking) in situations where vehicles and pedestrians may become occluded by otherobjects, or when tracking is momentarily lost. Benefits to Canada for the improved BriskVANTAGE softwareincludes higher quality and higher reliability data analytics for urban planners engaged in traffic monitoring andanalysis, leading to safer roads and improved traffic flow. Real-time systems can provide immediate signalingof significant traffic events as well as long-term monitoring of traffic and safety conditions and emergent trendsdue to changes in demographics and road construction.