In an interview with the Tucson commercial vehicle automatic driving technology will be the road as

In an interview with the Tucson: commercial vehicle automatic driving will be the road as soon as possible – Sohu of science and technology in September 19th, Tucson Internet announced that computer vision and deep learning algorithm they developed in automatic driving algorithm in public list KITTI and Cityscapes evaluation data sets were obtained excellent results. This is a domestic focus on computer vision and artificial intelligence research and development enterprise, Tucson CEO Chen Mo is a serial entrepreneur, a company he was acquired after he took over the Internet, tucson. Established in September 2015 the establishment of the Internet in Tucson, Dr. CTO Hou Xiaodi, the company in the United States set up a study on visual recognition of 20 people, including team members from UCLA, Kanai Ki Mellon, MIT, Nanyang Polytechnic University of chinese. At first, they developed advertising visual recognition technology, such as Sina micro-blog Tussan has developed a "knowledge artifact": as long as the user conveniently upload pictures of cars, micro-blog can identify the mainstream car brand, model, conducive to accurate advertising. In March of this year, Chen gave himself and the company a huge challenge, comprehensive transformation. The automatic driving Tucson from advertising began to transition the most Applied Research Institute of North American visual recognition in the field of advertising, through precise matching algorithm, can improve the conversion rate of advertising. But Chen Mo believes that the core competitiveness of advertising is the ability to obtain sales and resources, and computer vision can help optimize the role is indeed limited. It’s the problem in data fraud too much, such as Facebook, their DSP products met a surprising number of false advertising and bad robot traffic, although they do screening, but still not be defeated by the value of the flow of the scale, finally decided by the quality of the first advertisement. So technology companies with great effort to invest may not be able to bring significant growth in sales. Compared to advertising, the difficulty of driving is very large, the automotive industry chain is very long, it is also a matter of 5-10 years to landing. Tucson, how to do? Chen Mo said that the short-term goal is to automatically drive the Tucson L3 level, it will be achieved within 5 years, for commercial vehicles in designated sections (such as highway) to achieve automatic driving function. There are a lot of problems to make the driving on the road, the traditional visual recognition algorithm is difficult to identify pedestrians and bicycles, but deep learning can be. At the same time, deep learning can also predict the behavior of objects around. Chen silence example: no one next to the car on the alarm, it is also possible to predict the possibility of collision with the vehicle, the need to predict the trajectory of both." Therefore, a good performance of the camera is not enough, object detection and classification, landmark recognition, driving behavior and the behavior of the vehicle to adapt to the decision-making, each step requires the involvement of deep learning. Chen is very sure that the traditional algorithm of visual recognition is not feasible, the general front car anti-collision is not enough, it will be one or two years after the depth of learning to replace. The provision of automated driving capabilities for commercial vehicles to respond to infrequent contingencies is a training.相关的主题文章: