CN-121981539-A - Vehicle record supervision system

CN121981539ACN 121981539 ACN121981539 ACN 121981539ACN-121981539-A

Abstract

The invention relates to the technical field of garbage transportation supervision, in particular to a vehicle record supervision system, which comprises a dynamic data compensation module, a weighing anomaly identification module, a mixed loading identification module and a risk assessment module, wherein the dynamic data compensation module realizes vehicle identification through multi-source information fusion and evidence demonstration algorithm and dynamically compensates weighing information by combining vehicle type and speed data, the weighing anomaly identification module detects weighing curve anomalies based on a conditional random field model and generates scores by combining track compliance detection, the mixed loading identification module extracts multi-modal characteristics in a discharging process and identifies mixed loading behaviors through a hidden Dirichlet distribution model and chi-square inspection, and the risk assessment module takes chi-square statistics, track compliance scores and weighing anomaly scores as characteristics and estimates quantitative risk levels through local Fisher discrimination. The system monitors the whole process of the garbage transport vehicle, solves the problems of inaccurate weighing, difficult mixed loading identification, inaccurate risk assessment and the like, and improves the monitoring efficiency.

Inventors

WU JIE
LIU DAQUAN
GAO LIANJIE
HAN FEI
WANG NING
LI MIN
WANG XUAN

Assignees

青岛环境再生能源有限公司

Dates

Publication Date: 20260505
Application Date: 20260119

Claims (8)

1. The vehicle record supervision system comprises a dynamic data compensation module, a weighing abnormality identification module, a mixed identification module, a risk assessment module and a database, and is characterized in that: The dynamic data compensation module acquires wagon balance weighing information, vehicle identity information, vehicle running information and garbage type information of the garbage transport vehicle through each subunit, acquires a vehicle identification result through an evidence demonstration algorithm, and dynamically compensates the wagon balance weighing information based on the vehicle type identification result and speed detection; The weighing anomaly identification module generates actual operation events of each vehicle, the weighing curve is subjected to anomaly detection through the conditional random field model to obtain the weighing anomaly score of the wagon balance, and then the track compliance score is obtained through the track compliance detection model; The mixed loading recognition module extracts action feature vectors, garbage color texture features and spectrum fingerprint features in the unloading process of the vehicle, and recognizes garbage mixed loading through a hidden Dirichlet allocation model and chi-square inspection; the risk assessment module is characterized by chi-square statistics, track compliance scores and wagon balance weighing anomaly scores, and the risk degree of the current garbage truck is assessed through nuclear density estimation and a local Fisher discrimination algorithm.
2. The vehicle proposal supervision system according to claim 1, wherein the risk assessment module is characterized by chi-square statistics, track compliance scores and wagon balance weighing anomaly scores, and assesses the risk level of the current refuse vehicle by a nuclear density estimation and local Fisher discriminant algorithm, specifically: obtaining obtained chi-square statistics, track compliance scores and wagon balance weighing anomaly scores in actual operation events of the current garbage truck, extracting operation records of historical data, dividing the operation records into normal sample categories and risk sample categories based on the operation records, carrying out k neighbor searching on all samples to obtain neighbor samples, calculating local similarity of the samples and corresponding neighbor samples, obtaining intra-class local divergence matrixes and inter-class divergence matrixes based on the local similarity, establishing a generalized eigenvalue problem through the intra-class local divergence matrixes and the inter-class divergence matrixes, solving the problem to obtain a group of eigenvectors and eigenvalues, sorting the front d eigenvectors according to the eigenvalue size to obtain a projection matrix, and mapping the chi-square statistics, the track compliance scores and the wagon balance weighing anomaly scores through the projection matrix to obtain risk discrimination scores; The risk grade of the vehicle is divided based on the risk discrimination score, wherein the risk grade comprises a high-risk vehicle, a medium-risk vehicle and a low-risk vehicle, probability density functions of the risk grades are obtained through a kernel density estimation algorithm, posterior probability of the current risk discrimination score belonging to the risk grades is calculated based on the probability density functions, and the corresponding maximum probability risk grade is sent to a record supervision server for marking.
3. The vehicle record supervision system according to claim 1, wherein the mixed recognition module extracts motion feature vectors, garbage color texture features and spectrum fingerprint features of a vehicle unloading process, and the specific process is as follows: Extracting geometric and motion parameters of a tipping bucket area of the vehicle through a target detection algorithm, wherein the geometric and motion parameters comprise a change curve of a carriage inclination angle along with time, a tail gate opening angle, vertical displacement of the tail end of the carriage during dumping and the growth speed of a garbage outflow pixel area, arranging the geometric and motion parameters according to time sequence to obtain action time sequence characteristics, and encoding the action time sequence characteristics into a low-dimensional action characteristic vector through a time convolution network; The method comprises the steps of obtaining a frame sequence in a discharging process, separating a garbage area from the background, a vehicle and the ground of the frame sequence through semantic segmentation, extracting masks of the garbage area in each frame, fusing continuous frames to obtain a garbage pile area integral mask, converting the integral mask into an HSV space, obtaining color histograms, color mean values, variances and skewness characteristics of all channels based on the HSV space, integrating the color characteristics, extracting texture characteristics in the garbage pile area integral mask, including local binary pattern histograms and gray level symbiotic matrix characteristics, and splicing the color characteristics and the texture characteristics to obtain garbage color texture characteristic vectors; Extracting a plurality of sampling points from hyperspectral data of a garbage area, converting the sampling points into a spectrum curve, obtaining spectrum characteristics of the spectrum curve, including spectrum slope, ratio characteristics, spectrum valley and peak positions, matching the spectrum characteristics with garbage material categories of a database to obtain garbage spectrum categories, and counting the occurrence frequency of each garbage spectrum category discharged by the vehicle to obtain spectrum category frequency vectors.
4. The vehicle record supervision system according to claim 3, wherein the mixed loading identification module identifies the garbage mixed loading through a cryptodirichlet allocation model and a chi-square test, and the specific process is as follows: Dividing the low-dimensional motion feature vector into a plurality of motion mode codes by using clustering, clustering the garbage color texture feature vector into a plurality of visual appearance codes, clustering the spectrum class frequency vector into a plurality of spectrum class codes, setting the unloading process as a document, and setting the corresponding motion mode code, visual appearance code and spectrum class code as the word frequency vector of the document; Extracting a plurality of subjects associated with garbage categories with different components from the subject distribution feature vector, merging to obtain observation category frequency distribution, pre-counting garbage subject distribution mean values of the categories through a single garbage category vehicle sample, setting the garbage subject distribution mean values as expected distribution, calculating chi-square statistics based on the observation category frequency distribution and the expected distribution, marking the vehicle as a mixed vehicle if the chi-square statistics is greater than a preset chi-square threshold value, and marking the vehicle as a non-mixed vehicle if the chi-square statistics is greater than the preset chi-square threshold value.
5. The vehicle proposal supervision system according to claim 1, wherein the dynamic data compensation module obtains wagon balance weighing information, vehicle identity information, vehicle operation information and garbage type information of the garbage transport vehicle through each sub-unit, and specifically comprises: Each subunit comprises a wagon balance weighing information unit, a vehicle identification subunit, an operation state acquisition subunit and a garbage type unit, wherein the wagon balance weighing information unit acquires gross weight, tare weight, net weight, time stamp and weighing curve data when a vehicle enters and exits a field, the vehicle identification subunit respectively identifies license plates, vehicle type identification tags and vehicle identity information of vehicle-mounted terminals ID, the operation state acquisition subunit acquires vehicle entering time, queuing time, loading time, unloading time, departure time, track in the field and vehicle speed information in real time, and the garbage type unit acquires garbage type information through manual entry or two-dimensional code scanning.
6. The vehicle proposal supervision system according to claim 5, wherein the dynamic data compensation module obtains a vehicle identification result through an evidence demonstration algorithm, and dynamically compensates wagon balance weighing information based on the vehicle type identification result and speed detection, and the specific steps are as follows: Constructing an identification framework by using unique identifications of all vehicles in a vehicle record database as element IDs, binding the elements to correspond to preset confidence, respectively setting license plate identification, vehicle type identification tags and vehicle terminal IDs as evidence sources, calculating basic probability distribution of each evidence source to obtain license plate probability distribution, vehicle type probability distribution and vehicle probability distribution, fusing all probability distribution by using a evidence demonstration algorithm synthesis rule to obtain fused probability distribution, and taking the element with the largest fused probability distribution as a vehicle identification result; Dividing the current garbage transport vehicle type according to the vehicle record information and the vehicle type recognition result to obtain vehicle type parameters, wherein the vehicle type parameters comprise the dead weight, rated load, wheel track and wheel base of the vehicle, calculating inertia coefficients based on the vehicle type parameters, extracting instantaneous acceleration and original weighing values in the weighing process of the vehicle, and calculating compensated weighing data through a compensation function.
7. The vehicle record supervision system according to claim 1, wherein the weighing anomaly identification module generates actual operation events of each vehicle, and performs anomaly detection on a weighing curve through a conditional random field model to obtain a wagon balance weighing anomaly score, which specifically comprises: Correlating wagon balance weighing information, vehicle running information and vehicle record information to generate actual operation events of the vehicle, wherein the actual operation events comprise operation IDs, correlated vehicle IDs, entrance weighing record IDs, exit weighing record IDs, running state records of correlated vehicle numbers, and correlated garbage types and source information; Abstracting the full life cycle of a weighing curve into a discrete finite state set, including a blank weight, a normal rising edge, a normal falling edge, a jump weight rising edge and a jump weight stabilizing section, acquiring observation characteristics corresponding to the discrete finite state set, constructing a multidimensional observation sequence corresponding to the finite state set, including an instantaneous weight value, a peak value mark, a stabilizing section duration and a first-order difference, and carrying out joint modeling on the finite state set and the observation sequence through a linear chain member random field to obtain a conditional random field model; the method comprises the steps of obtaining an input weighing curve observation sequence to be measured, solving an optimal state sequence by adopting a Viterbi algorithm, counting the proportion of the accumulated time length of an abnormal state to the total effective time length of the weighing curve through the optimal state sequence, obtaining an abnormal state proportion score through nonlinear mapping, counting the legal proportion of abnormal state transition based on a transfer characteristic function to obtain an abnormal transfer rationality score, counting the state characteristic function mean value corresponding to the abnormal state to obtain an abnormal state matching degree score, and carrying out weighted summation on the abnormal state proportion score, the abnormal transfer rationality score and the abnormal state matching degree score to obtain the wagon balance weighing exception score.
8. The vehicle docket supervisory system according to claim 7, wherein the weighing anomaly identification module obtains a track compliance score via a track compliance detection model, comprising the specific steps of: Mapping track points to a road network based on a GPS track of a vehicle and a recorded line set to obtain a corresponding road segment sequence, marking the road segment sequence as a boundary crossing distance and a boundary crossing duration time if the road segment sequence has a preset distance threshold value exceeding the recorded line range or deviates from a recorded time window, normalizing the boundary crossing distance and the boundary crossing duration time, and weighting calculation to obtain a track compliance score.

Description

Vehicle record supervision system Technical Field The invention belongs to the technical field of garbage transportation supervision, and particularly relates to a vehicle record supervision system. Background The garbage transportation supervision is a key link of environment management and city management, is directly related to garbage disposal compliance, environment safety and transportation efficiency, and along with the promotion of garbage classification policies, the whole-flow supervision requirement on the garbage transportation vehicle is increasingly urgent; The conventional wagon balance weighing is easily affected by factors such as vehicle type difference, weighing speed, inertia impact and the like, has larger data deviation and lacks dynamic compensation, the mixed loading identification relies on manual check, the supervision efficiency is low, the characteristics such as actions, vision, spectrums and the like of the unloading process cannot be effectively extracted, the mixed transportation behavior cannot be effectively judged, the track supervision can only be simply positioned, quantitative evaluation on line deviation or time window violations is lacking, compliance judgment is rough, and the risk evaluation does not integrate multidimensional characteristics such as mixed loading, weighing abnormality, track violations and the like, and the evaluation result is one-sided, so that the data of each supervision link are isolated, closed-loop management is not formed, and the supervision efficiency is affected; thus, there is a need for a vehicle docket supervisory system that addresses the above-described issues. Disclosure of Invention In order to solve the technical problems set forth in the background art, the invention provides a vehicle record supervision system. The aim of the invention can be achieved by the following technical scheme: The invention provides a vehicle record supervision system which comprises a dynamic data compensation module, a weighing abnormality identification module, a mixed loading identification module, a risk assessment module and a database. The dynamic data compensation module obtains wagon balance weighing information, vehicle identity information, vehicle running information and garbage type information of the garbage transport vehicle through each subunit, obtains a vehicle identification result through an evidence demonstration algorithm, and dynamically compensates the wagon balance weighing information based on a vehicle type identification result and speed detection, wherein the wagon balance weighing information comprises the following specific steps of: each subunit comprises a wagon balance weighing information unit, a vehicle identification subunit, an operation state acquisition subunit and a garbage type unit, wherein the wagon balance weighing information unit acquires gross weight, tare weight, net weight, time stamp, weighing curve data and the like when a vehicle enters and exits a field, the vehicle identification subunit respectively acquires vehicle identity information such as license plate identification, vehicle type identification tag, vehicle terminal ID and the like, the operation state acquisition subunit acquires vehicle entering time, queuing time, loading time, unloading time, departure time, in-field track, vehicle speed information and the like in real time, and the garbage type unit acquires garbage type information in a manual entry mode or a two-dimensional code scanning mode and the like; Building an identification framework by using unique identifications of all vehicles in a vehicle record database as element IDs Element binding corresponds to preset confidence, n is the total number of vehicles, license plate recognition, vehicle type recognition tags and vehicle-mounted terminal IDs are respectively set as evidence sources, basic probability distribution of each evidence source is calculated, and license plate probability distribution is obtainedProbability distribution for vehicle modelOn-vehicle probability distributionThe calculation logic is as follows:、、 wherein In order to match the identification confidence coefficient with the unique identification of the vehicle, it is to be noted that the step mainly adopts the identification confidence coefficient of the license plate corresponding to the vehicle identification which is output by the license plate probability distribution, namely, the corresponding confidence coefficient of the matching identification is given, the other identifications are 0, when the vehicle type evidence source identification is successful, the basic probability distribution of the corresponding vehicle identification is set as 1, the basic probability distribution of the whole identification frame is given as 1 when the identification fails, the vehicle-mounted evidence source is the same, and the probability distribution is fused through the evidence demonstration algorithm synt