Image target detection method, device and camera equipment

Abstract

The application discloses an image target detection method, device and imaging equipment, which are used to improve the existing detectors so that they can flexibly adapt to different detection ranges and meet different product requirements. The method of the present application includes: determining the current application scene of the preset detector, and obtaining the corresponding configuration file according to the application scene; obtaining the image to be detected; reading the positional relationship from the configuration file, and the positional relationship is the Presetting the positional relationship between the detector sub-region and the image to be detected in the application scene; dividing the image to be detected according to the positional relationship to obtain a sub-image corresponding to the sub-region; through the preset It is assumed that the detector performs target detection on the sub-image to obtain a detection result; the detection result is processed to obtain the target detection result of the image to be detected.

Description

technical field [0001] The present application relates to the technical field of image detection, and in particular to an image object detection method, device and camera equipment. Background technique [0002] Nowadays, target detection applications based on face detection, head detection, pedestrian detection, and vehicle detection are very common. Target detection technology is widely used in consumer electronics, smart terminals, commercial displays, transportation, security and other fields. Most target detection algorithms are based on deep learning, and the corresponding deep learning models are usually trained based on specific data sets or specific scenarios. Once deployed, their specifications (such as detection accuracy, detection distance) and the scenarios they can handle are determined. of. In actual application scenarios, the same company may have multiple products or multiple functions of the same product that require the same target detection algorithm, bu...

AI Technical Summary

Problems solved by technology

However, the former is time-consuming and laborious, with high R&D costs and a long cycle, while the latter is extremely difficult to train and may not even be feasible. Even if it can meet the effect, its performance is often unbearable for most mobile devices, especially low-end embedded platforms.

[0004] To sum up, if the product demand changes, it will take time and effort to select and train a new detector. For the demand of a larger scale range, it is difficult to train a new detector. Even if it is trained, the computational overhead will inevitably be very high. Compared with the original detector, the single frame detection time will be greatly increased. For the low-end embedded platform, the corresponding calculation speed and detection frame rate will be greatly reduced.

Images