Researchers Submit Patent Application, “Method, Storage Medium And Electronic Device For Detecting Vehicle Crashes”, for Approval (USPTO 20190050711)

2019 FEB 28 (NewsRx) -- By a News Reporter-Staff News Editor at China Business Day -- From Washington, D.C., NewsRx journalists report that a patent application by the inventors YAO, Jian (Shenyang, China); Zhang, Michael (Shenyang, China); Xu, Lili (Shenyang, China), filed on December 29, 2017, was made available online on February 14, 2019.

The patent’s assignee is Neusoft Corporation (Shenyang, People’s Republic of China).

News editors obtained the following quote from the background information supplied by the inventors: “Automatic detection of vehicle crashes is beneficial to timely notifying crash accidents to relevant personnel and organizations, including first-aid personnel, family members, team principals and insurance companies. On the other hand, timely detection of crash accidents is also beneficial to investigating the accidents.

“In some relevant technologies, vehicle crashes are automatically detected directly using crash detection hardware sensors. In some other relevant technologies, operational data of vehicles are acquired using vehicle-mounted sensors or mobile sensors, and feature values are calculated via the methods of integration, difference and the like according to the sensor data. Then, thresholds are calculated via these feature values to determine whether crashes happen.”

As a supplement to the background information on this patent application, NewsRx correspondents also obtained the inventors’ summary information for this patent application: “In order to solve the problems in relevant technologies, the present disclosure is aimed at providing method, apparatus, electronic device and vehicle for detecting vehicle crashes.

“In a first aspect, the present disclosure provides a method for detecting vehicle crashes, including:

“acquiring state information of a target vehicle; and

“determining an event type of the target vehicle according to the state information and a trained convolutional neural network, the event type being any of the following types: a crash event, a near crash event and a baseline event.

“In a second aspect, the present disclosure provides an apparatus for detecting vehicle crashes, including:

“an acquisition module, used for acquiring state information of a target vehicle; and

“a determination module, used for determining an event type of the target vehicle according to the state information and a trained convolutional neural network, the event type being any of the following types: a crash event, a near crash event and a baseline event.

“In a third aspect, the present disclosure provides a computer readable storage medium, storing a computer program which, when executed by a processor, performs the steps of said method.

“In a fourth aspect, the present disclosure provides an electronic device, including:

“the computer readable storage medium in said third aspect; and

“one or more processors, used for executing the program in the computer readable storage medium.

“In a fifth aspect, the present disclosure provides a vehicle, including:

“the computer readable storage medium in said third aspect; and

“one or more processors, used for executing the program in the computer readable storage medium.

“In said technical solutions, the event type of the vehicle is determined using the trained convolutional neural network, so that the accuracy is high; and near crash events can be detected, thus, when a near crash event is detected, the driver can be further alerted or an evading operation (braking, abrupt turning, or the like) can be directly performed on the vehicle, so that safety is improved and the safety of the driver and passengers is guaranteed.

“Other features and advantages of the present disclosure will be described in detail in the following specific embodiments.”

The claims supplied by the inventors are:

“1. A method for detecting vehicle crashes, comprising: acquiring state information of a target vehicle; and determining an event type of the target vehicle according to the state information and a trained convolutional neural network, the event type being any of the following types: a crash event, a near crash event and a baseline event.

“2. The method of claim 1, further comprising: acquiring a training sample, wherein the training sample comprises: multiple pieces of time series data and event type tags corresponding to each piece of time series data, wherein each piece of time series data comprises state information of the vehicle recorded by at least one sensor according to time; training a convolutional neural network according to the training sample and a training termination condition; when the training is terminated, acquiring parameter information of the convolutional neural network to be trained, wherein the parameter information at least comprises: weights of a convolution layer, biases of the convolution layer, weights of a pooling layer, biases of the pooling layer, weights of a fully connected layer, biases of the fully connected layer, number of convolution layers, size of the convolution kernel of each convolution layer, number of pooling layers, size of each pooling layer, number of fully connected layers and size of each fully connected layer; and constructing the convolutional neural network according to the parameter information.

“3. The method of claim 2, wherein the step of acquiring a training sample comprises: acquiring time series data with an event type tag from at least one sensor; merging the time series data with event type tags from different sensors based on timestamps; and determining the merged time series data with event type tags as the training sample.

“4. The method of claim 3, wherein the step of merging the time series data with event type tags from different sensors based on timestamps comprises: among the time series data with event type tags from different sensors, segmenting the time series data of the same event type into multiple pieces of time series data based on a minimum time window corresponding to the event and a preset time window moving amount; and merging the segmented time series data from different sensors based on timestamps.

“5. The method of claim 4, wherein the step of merging based on timestamps comprises: when the timestamps of the time series data from different sensors are different, performing linear interpolation on the time series data with a low sampling frequency; and merging the time series data after linear interpolation to obtain time series data to be sampled; the step of determining the merged time series data with event type tags as the training sample comprises: determining time series data sampled from the time series data to be sampled at a preset sampling frequency and the corresponding event type tags thereof as the training sample.

“6. The method of claim 2, further comprising: discarding a preset number of neurons in the fully connected layer at each iteration.

“7. The method of claim 2, further comprising: acquiring a test sample, wherein the test sample comprises state information of a vehicle to be tested and an event type tag corresponding to the state information; inputting the state information of the vehicle to be tested into a convolutional neural network constructed with the parameter information to acquire an event type of the vehicle to be tested; and when the acquired event type of the vehicle to be tested is not accordant with the event type tag, retraining the convolutional neural network according to the training sample to update the parameter information.

“8. The method of claim 1, wherein the state information of the vehicle comprises: speed, acceleration in the X direction, acceleration in the Y direction, acceleration in the Z direction, angular speed in the X direction, angular speed in the Y direction and angular speed in the Z direction.

“9. The method of claim 1, wherein the step of determining an event type of the target vehicle according to the state information and a trained convolutional neural network, comprises: preprocessing the state information; and determining an event type of the target vehicle according to the preprocessed state information and a trained convolutional neural network.

“10. The method of claim 9, wherein the state information is the time series data from different sensors recorded according to time; the step of preprocessing the state information comprises: merging the time series data from different sensors recorded according to time, based on timestamps.

“11. The method of claim 1, further comprising: when the event type of the target vehicle is the near crash event, outputting alerting information; and when the event type of the target vehicle is the crash event, outputting alarm information.

“12. A computer readable storage medium, storing a computer program which, when executed by a processor, performs the steps of the method of claim 1.

“13. An electronic device, comprising: the computer readable storage medium of claim 12; and one or more processors, used for executing the program in the computer readable storage medium.”

For additional information on this patent application, see: YAO, Jian; Zhang, Michael; Xu, Lili. Method, Storage Medium And Electronic Device For Detecting Vehicle Crashes. Filed December 29, 2017 and posted February 14, 2019. Patent URL: http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PG01&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.html&r=1&f=G&l=50&s1=%2220190050711%22.PGNR.&OS=DN/20190050711&RS=DN/20190050711

(Our reports deliver fact-based news of research and discoveries from around the world.)