Research On Vehicle Logo And Seat Belt Recognition Based On Image Processing And Machine Learning

Target detection and recognition in digital images processing is the process of converting image signals into digital signals and using image processing related methods to segment the target of interest from the images and solving the target's feature extraction and recognition.It is one of the most important research fields of artificial intelligence.With the widespread use of surveillance cameras,the research on specific target recognition in video and images has developed rapidly in the field of intelligent surveillance systems.Among them,the vehicle characteristics of roadside bayonet video images and the identification of illegal behaviors of occupants are the research hotspots in recent years.At present,there are many researches on machine vision-based vehicle type identification and driver's violation of regulations recognition,but the actual application is not mature enough.The types of models that can be identified by the existing system are few,and the accuracy of violations identification by drivers is not high.It is still necessary to further study more effective and feasible methods.This paper uses adaptive scaling sliding window and noise filtering masks technology to identify the vehicle logo for the first time.Than extract the HOG features of the matching region images corresponding to the first 6 recognition results with the highest matching rate.Finally,uses the support vector machine(SVM)to train and classify the vehicle logos for the second times.At the same time,the original slash detection operators was used to achieve accurate identification of seat belt.The main contributions of this study are as follows:1.A method of vehicle logo recognition based on adaptive scaling sliding window and mask technology is proposed.In this paper,an adaptive scale sliding window is used to achieve the accurate positioning of the vehicle logo,and the scaling adaptive mask technique is used to effectively eliminate the background noise around the vehicle logo.On this basis,the template matching and the weighted matching algorithm are used to identify the target vehicle logos.2.A secondary vehicle identification method based on SVM is proposed.By extracting the HOG features of the logos and using the two-category of SVM algorithm to train the features,the trained SVM classifier is used to recognition the first six recognition results with thehighest matching rates.The double recognition ensures the accuracy.3.In order to reduce the amount of calculation of sliding template matching,effective measures have been taken: after the vehicle logo is positioned coarsely,we categorize the vehicle logo initially by judging the type of the inlet texture.So that,the amount of calculation of template matching can be reduced.What's more,the template matching stopping condition is set.During the template matching process,If the matching rate is greater than the threshold,than the program stop adaptively and output the recognition result.4.A specific detection operator for the seat belt recognition is proposed to detect the unique characteristics of the seat belt.The detection operator can extract the slants of the seat belt under the condition with high noise,which as a supplementary feature of the target vehicle for further screening of targets.5.The case that partial region of the seat belt was covered by the driver's collar or affected by the reflective noise,the edge of the seat belt extracted from the image would be fractured.In order to extract the whole edge,we calculate the gradient of the two lines and the line that connected by the bottom point of the lower line and the top point of the higher line,when the value of the three gradients are almost equivalent,them connect the two lines.This method improves the accuracy rate of the seat belt recognition.