Automatic driving target detection system and method based on deep learning and binocular camera shooting

Abstract

The invention provides an automatic driving target detection system and method based on deep learning and binocular camera shooting. The problems of poor target detection accuracy and poor object positioning accuracy in existing automatic driving are mainly solved. Complex road conditions cannot be dealt with. The implementation scheme is as follows: the method comprises the following steps: 1, implementing; the binocular camera is used for collecting images and carrying point cloud information. The signals are transmitted to an image processing module through a data line for real-time detection; and the image processing module is a detection module constructed after training data by using a deep neural network, the image processing module processes the acquired images to realize positioning, recognition and classification, and finally, the result output module splices classification results into a video and outputs the video to complete detection of road conditions in automatic driving. The method improves the target recognition precision, solves a problem that automatic driving cannot adapt to complex road conditions, and can be used for an automatic driving and driving active safety system.

Description

technical field [0001] The invention belongs to the field of artificial intelligence technology, and further relates to the field of target recognition and detection, specifically an automatic driving target detection system and method based on deep learning and binocular imaging, which can be applied to automatic driving and driving active safety systems. Background technique [0002] Artificial intelligence technology is currently the most novel and potential research field in the world. Among them, artificial intelligence has many application scenarios, including natural language processing, computer vision, speech recognition, etc. Since Google's aphaGo artificial intelligence defeated the world champion Lee Sedol in the field of Go in 2017, artificial intelligence has become a topic that has been talked about on the streets for a while. With the rapid development of artificial intelligence technology in recent years, more and more related applications have entered peopl...

AI Technical Summary

Problems solved by technology

It uses an image acquisition module and an infrared polarized light imaging module to collect images, and then uses the CNN neural network of deep learning technology to learn and train the images to complete the detection and identification, but there are still the following deficiencies: (1) The system uses an ordinary camera with only a single lens, so the range of image acquisition is narrow, and it is impossible to observe the road conditions more comprehensively. The blind area of ​​the viewing angle is large, and there is no comparison object. If the camera fails, it may cause the system to fail; (2) The system uses the CNN network to detect the images collected by the camera. This network can only be used to identify the object classification in the image, and cannot determine the position of the object in the image and mark the object, so it cannot be applied to automatic driving. In the middle, and the network has appeared for a long time, the effect in target detection is not ideal, especially for more complex scenes, the detection results are poor, and it is impossible to accurately distinguish each object in the complex scene, which is often faced by automatic driving. For complex road conditions, it is incompetent

Images