185results about How to "Reduce misidentification" patented technology

A feature extraction method includes: the step of grouping a cluster of features, in which an internal of the respective features is less than or equal to a predetermined grouping interval, to form a feature group, for a plurality of features of which feature information including at least information of a position and a feature type is included in a predetermined feature information storage unit; the step of excluding the feature, not suitable for use in an image recognition process of the feature with respect to image information, from the cluster of the features within the feature group; and the step of extracting a part or all of one or more of the features within the feature group remaining as a result of the exclusion step as a target feature suitable for the use in the image recognition process.

A game system includes a game apparatus and a controller. The controller is furnished with an acceleration sensor for detecting accelerations in at least two axis directions. Game processing corresponding to the kind of an acceleration input by means of the controller is executed. For determining the kind, reference timing when acceleration in a first-axis direction is below a threshold value to take on a minimum value is detected. Then, it is determined whether or not an angle between acceleration change vectors before and after the reference timing is equal to or more than a predetermined angle. When the angle is not equal to or more than the predetermined angle, it is determined that the acceleration input is an acceleration input in any one of the two-axis directions, and when the angle is equal to or more than the predetermined angle, it is determined that the acceleration input is an acceleration input in a direction including the two-axis directions as components.

In conventional systems using an onboard camera disposed rearward of a vehicle for recognizing an object surrounding the vehicle, the object is recognized by the camera disposed rearward of the vehicle. In the image recognized by the camera, a road surface marking taken by the camera appears at a lower end of a screen of the image, which makes it difficult to predict a specific position in the screen from which the road surface marking appears. Further, an angle of depression of the camera is large, and it is a short period of time to acquire the object. Therefore, it is difficult to improve a recognition rate and to reduce false recognition. Results of recognition (type, position, angle, recognition time) made by a camera disposed forward of the vehicle, are used to predict a specific timing and a specific position of a field of view of a camera disposed rearward of the vehicle, at which the object appears. Parameters of recognition logic of the rearwardly disposed camera and processing timing are then optimally adjusted. Further, luminance information of the image from the forwardly disposed camera is used to predict possible changes to be made in luminance of the field of view of the rearwardly disposed camera. Gain and exposure time of the rearwardly disposed camera are then adjusted.

A method for speech recognition includes providing a source of geographical information within a vehicle. The geographical information pertains to a current location of the vehicle, a planned travel route of the vehicle, a map displayed within the vehicle, and / or a gesture marked by a user on a map. Words spoken within the vehicle are recognized by use of a speech recognition module. The recognizing is dependent upon the geographical information.

The invention relates to an augmented reality-based Chinese Pinyin combination recognition and reading learning system. The augmented reality-based Chinese Pinyin combination recognition and reading learning system includes recognition and reading Pinyin cards with pinyin pictures, a Pinyin picture image recognition module, a Pinyin combination processing module, a Pinyin recognition and reading module and a multimedia module; the Pinyin picture image recognition module calls the Pinyin combination processing module to process obtained Pinyin data when encountering a plurality of Pinyin cards; the output end of the Pinyin picture image recognition module is connected with the input end of the Pinyin recognition and reading module, so that multimedia unit marks formed by initial consonant combinations and final vowel combinations of the Pinyin can be obtained from the recognition and reading Pinyin cards with the Pinyin pictures and are recognized; the output end of the Pinyin recognition and reading module is connected with the input end of the multimedia module, so that specific pronunciations of Pinyin combinations and multimedia images related to the pronunciations can be outputted; the recognition and reading Pinyin cards with the Pinyin pictures are cards printed with Pinyin initial consonant images or final vowel images; and each card is printed with a Pinyin picture with unique identification features.

A sheet-convey device, includes a holder, a convey member, and a recognizer. On the holder, a first sheet having a first width and a second sheet having a second width larger than the first width are placeable. The convey member conveys a sheet placed on the holder in a predetermined direction. The recognizer recognizes a width of the sheet and includes first and second sensors for sensing whether a part of the sheet is present at respective first and second positions of the holder. The first position is located inside of the first width in the widthwise direction with the first sheet placed on the holder, while the second position is located outside of the first width and inside of the second width with the second sheet placed on the holder. The second position is located upstream of the first position in the predetermined direction.

A sheet-convey device, including: a holder on which a first sheet having a first width and a second sheet having a second width larger than the first width are placeable at a predetermined setting position; a convey member which conveys a sheet placed on the holder in a predetermined direction perpendicular to a widthwise direction of the sheet; a recognizer which recognizes a width of the sheet; wherein the recognizer includes first and second sensors for sensing whether a part of the sheet is present at respective first and second positions of the holder; wherein the first position is located inside of the first width in the widthwise direction with the first sheet placed on the holder, while the second position is located outside of the first width with the first sheet placed on the sheet holder and inside of the second width with the second sheet placed on the holder; and wherein the second position is located upstream of the first position in the predetermined direction.

The embodiment of the application discloses a training method and device for an image recognition model, which obtains a plurality of application images of an application and determines the color distribution characteristics corresponding to each application image respectively, and performs the image clustering according to the color distribution characteristics of the plurality of application images to obtain a plurality of image sets. In the same set of clustered images, the color distribution features of the applied images are similar to each other, which generally accords with the first threshold, so that the applied images in the same set of images are likely to belong to the same set of application scenes. Based on the image set obtained by the clustering mentioned above, the image recognition model is trained according to a class of image sets, so that the image recognition model can be used to recognize the application images in the application scenes corresponding to this class of image sets with high accuracy, thus reducing the possibility of false recognition results.

The invention provides a target identification method based on a deep learning network and a hot-line robot. Weights and offsets on each layer in a convolution neural network are initialized; two-dimensional image data is input for convolution to obtain C1-layer data; the maximum pooling treatment is performed for the C1-layer data to reduce the magnitude of the data to obtain S2-layer data; the convolution is performed for the S2-layer data to obtain C3-layer data; the maximum pooling treatment is performed for the C3-layer data to further reduce the magnitude of the data to obtain S4-layer data; the convolution is performed for the S4-layer data to obtain C5-layer data; the C5-layer data serves as input to totally connect with a H6 layer, and an output layer is totally connected with the H6 layer to output learning results of target characteristics; and the weights and the offsets are finely adjusted by applying a back propagation algorithm to finish the network learning. The method is high in target identification rate and excellent in robustness under complicated background.

A pressurized flow of breathable gas is delivered to the airway of a subject in accordance with a therapy regimen. One or more fluid parameters of the pressurized flow of breathable gas are adjusted based on the therapy regimen. The therapy regimen dictates that such adjustments be made based on the respiratory state of the subject. Transitions in respiratory state are identified without relying on measurement or estimation of flow at or near the airway of subject. Transitions in respiratory state are identified based on changes in the first time derivative of flow at or near the airway of the subject. In one embodiment, an effort parameter is determined that approximates the second time derivative of flow. Based on comparisons of the effort parameter to a dynamic threshold, transitions in respiratory state are identified.

The invention discloses a license plate character recognition method based on data enhancement and data generation. According to the method, a simulated license plate picture can be randomly generated; after the simulated license plate picture is obtained by the image generator of a cyclic adversarial network, and the simulated license plate picture is mixed with a simulated license plate pictureand an enhanced data picture to generate training data, so that optimization training can be trained on a license plate recognition network; and finally, the trained recognition network is subjected to rapid license plate recognition. According to the method of the invention, a license plate can be correctly and rapidly identified under the influence of low light, low resolution, motion blur and other severe conditions. Meanwhile, a large amount of data needed by neural network training is obtained in a data enhancement and data generation mode, and a large amount of manually-labeled data is not needed.

A signboard apparatus includes: light panel apparatuses arranged two-dimensionally, and each having a transmission function and a light-emitting face for visible light communication; and a control apparatus that performs, with respect to the light panel apparatuses, exclusive control of the transmission function. The light panel apparatuses are arranged as a collective signboard.