FIELD OF INVENTION
[0001] The present invention relates to classification of moving ground targets using Doppler Radar.

BACKGROUND
[0002] In a conventional method for recognizing a radar target, a Doppler feature vector is extracted from the return signal and is used for classifying the target into a particular class using Hidden Markov Modeling (HMM). In this method, each Doppler feature vector is assigned an occurrence probability corresponding to each class. Multiplying all the occurrence probabilities together for a sequence of Doppler feature vectors gives the overall probability of a particular class. The overall probabilities are found for all the classes and the target class is the one which yields the highest overall occurrence probability. Using the available training data for target classes, HMM estimates the state or probability distributions and transition probabilities of Doppler feature vectors for the different target classes. The estimated HMM parameters are used during target recognition.
[0003] In another conventional method for identifying moving objects using Doppler Radar, the unknown target is identified using a statistical classifier. This method explores the fact that the reflected signal from the target received by the Radar has a range of frequencies different from that of the transmitted signal. The power spectral measurements corresponding to each segment are compressed either using a logarithmic or cube root function to extract general features for classification. The extracted features are subjected to Non-linear Discriminant Analysis or Singular Value Decomposition for feature dimensionality reduction. Different statistical classifiers stated as used for target classification in this method include Neural Network, Gaussian Mixture Model, Hidden Markov Model and Support Vector Machine.
[0004] In yet another conventional method of Recognizing a Radar target object type, a plurality of Radar Candidate Signatures (RCS) is stored for different classes of targets. The measured time versus frequency diagram of a target is compared with the stored plurality of predicted time versus frequency diagrams to identify the closest matching target class. The predicted reflected echo is computed by convolving the transmitted pulses with the known target impulse response. The knowledge of target range information, heading, speed information and aspect angle is used in this method.
[0005] In yet another conventional method of classifying patterns, Radar targets are classified using Wavelet Transform. The signal received by the Radar is sampled for analysis. The extracted wavelet feature vector is transformed to a reduced dimensional space. The method uses a quadratic classifier for classifying the dimension reduced feature vector into different target classes.
[0006] There is still a need for effective method of classifying moving ground targets using the return signals of a Doppler Radar.

SUMMARY
[0007] This summary is provided to introduce concepts related to a radar spectrum characterization method. This summary is neither intended to identify essential features of the present invention nor is it intended for use in determining or limiting the scope of the present invention.
[0008] In an embodiment of the present invention, a radar spectrum characterization method for determining valid peaks in the spectrum of a radar return signal is provided. The method includes receiving the radar return signal at a processor. The processor computes a Fast Fourier Transform (FFT) of the received radar return signal to obtain a spectrum. The processor quantizes the spectrum into a plurality of amplitude levels. The processor computes a lowest frequency (fL) of a rightmost peak of the quantized spectrum. The processor determines a highest frequency (fH) of the rightmost peak of the quantized spectrum and a lowest quantized level (LL) and a highest quantized level (LH). The processor computes a maximum amplitude (Amax), a minimum amplitude (Amin), a top frequency width, and a bottom frequency width. The processor updates the lowest frequency (fL) as a processing end frequency after determining whether a peak is valid peak in the radar return signal.
[0009] In an exemplary embodiment of the present invention, the step of quantizing includes dividing the spectrum into the plurality of levels. Thereafter, the processor determines a number of level transitions including rising and falling edges for each of the plurality of levels within a predefined frequency range.
[0010] In another exemplary embodiment of the present invention, the processor sorts a plurality of rising edge frequencies in the predefined frequency range. Thereafter, the processor determines the peak to be valid peak if the number of level transitions from the lowest quantized level (LL) to the highest quantized level (LH) between a processing sorted frequency and a processing end frequency is at least two.
[0011] In another exemplary embodiment of the present invention, the processor re-estimates the lowest frequency (fL), the highest frequency (fH), and the highest quantized level (LH) based on the centre frequency of the peak. Thereafter the processor merges the peak with adjacent peak when a centre frequency of the adjacent peak lies between the re-estimated lowest frequency (fL) and the re-estimated highest frequency (fH).
[0012] In another exemplary embodiment of the present invention, the radar return signal is classified as noise if a maximum absolute value of the spectrum is less than a predefined threshold (ATh).
[0013] In another exemplary embodiment of the present invention, the processor determines a number of transitions in the spectrum. Thereafter, the processor classifies the spectrum as noise if the number of transitions in the spectrum exceeds a predefined threshold number.
[0014] In another exemplary embodiment of the present invention, the radar return signal is classified as a moving vehicle if at least one of the following conditions is true: (i) the spectrum has at least one peak covering a plurality of levels and a centre frequency (fC) is greater than a predefined threshold frequency, (ii) the spectrum has at least two peaks having more levels than a predefined number of levels and corresponding centre frequencies (fC) are greater than the predefined threshold frequency, and(iii) the spectrum has a peak covering a plurality of levels and has a small initial spectrum width and a small peak frequency width.
[0015] In another exemplary embodiment of the present invention, the radar return signal is classified as a moving pedestrian if the spectrum has high initial spectrum width and does not have a peak at frequencies greater than a predefined threshold frequency.
[0016] In another exemplary embodiment of the present invention, the radar spectrum characterization method is executed for every valid peak of the radar return signal.

BRIEF DESCRIPTION OF ACCOMPANYING DRAWINGS
[0017] The detailed description is described with reference to the accompanying figures.
[0018] Fig. 1 illustrates the various functional blocks involved in the proposed moving target classification method in accordance with an embodiment of the present invention.
[0019] Figs. 2A-2D illustrate a flow chart describing the steps involved in the characterization of the spectrum of the input signal in accordance with an embodiment of the present invention.
[0020] Figs. 3A-3G illustrate a flow chart describing the overall target recognition scheme in accordance with an embodiment of the present invention.
[0021] Fig. 4 illustrates a radar spectrum characterization system in accordance with an embodiment of the present invention.
[0022] It should be appreciated by those skilled in the art that any block diagram herein represent conceptual views of illustrative systems embodying the principles of the present invention. Similarly, it will be appreciated that any flow chart, flow diagram, and the like represent various processes which may be substantially represented in computer readable medium and so executed by a computer or processor, whether or not such computer or processor is explicitly shown.

DETAILED DESCRIPTION
[0023] The various embodiments of the present invention provide a method for classifying moving ground targets using Doppler Radar.
[0024] In the following description, for purpose of explanation, specific details are set forth in order to provide an understanding of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced without these details. One skilled in the art will recognize that embodiments of the present invention, some of which are described below, may be incorporated into a number of systems.
[0025] However, the systems and methods are not limited to the specific embodiments described herein. Further, structures and devices shown in the figures are illustrative of exemplary embodiments of the present invention and are meant to avoid obscuring of the present invention.
[0026] Furthermore, connections between components and/or modules within the figures are not intended to be limited to direct connections. Rather, these components and modules may be modified, re-formatted or otherwise changed by intermediary components and modules.
[0027] References in the present invention to “embodiment” or “embodiment” mean that a particular feature, structure, characteristic, or function described in connection with the embodiment or the embodiment is included in at least one embodiment or embodiment of the invention. The appearances of the phrase “in an embodiment” in various places in the specification are not necessarily all referring to the same embodiment.
[0028] The objective of the present invention is to develop a method for classifying moving ground targets using Doppler Radar. The signal from the Doppler Radar hits the moving target and returns to the Radar. The return signal has characteristics of the moving target incorporated in it. The return signal is sampled at 5000 Hz using a digitizer. 2048 samples are collected from the digitizer to constitute a data bin and subjected to classification to give a near real-time output and in turn, the classification result is updated twice every second. In an example, the return signal includes the reflected signal, i.e., the signal reflected from the moving target.
[0029] In an embodiment of the present invention, the Radar signal is classified into one of the three classes namely Moving Pedestrian, Moving Vehicle or Clutter. When the signal from Radar hits a moving target, the return signal changes based on the aspect angle, moving speed and the rotating parts of the target. The moving arms and legs of a walking pedestrian induce changes in the return signal and most of the energy is concentrated in the lower frequency band of the spectrum. The rotating wheels or chains of moving vehicle induce changes and exhibit peaks in the spectrum in accordance to the speed of the vehicle. Clutter has less energy and has no proper pattern in the spectrum.
[0030] In an exemplary embodiment of the present invention, the spectrum of the return signal samples stored in the data bin is characterized. To characterize the spectrum, the FFT of the input is computed, smoothed using a moving average filter and then quantized into a fixed ‘N’ number of levels in the amplitude axis. Rising edge and falling edge level transitions are identified for all the levels for analysis. When noise gets embedded with the target’s signature, there are many level transitions in the lower levels and need to be ignored while characterizing the spectrum. Therefore, the lowest processing level is computed for further analysis.
[0031] In an exemplary embodiment, various parameters of a peak such as Level-Lowest(LL), Level-Highest(LH), centre frequency(fC), lowest frequency(fL), highest frequency(fH), top frequency width & bottom frequency width are computed for all the peaks in the spectrum one by one starting with the rightmost peak of the spectrum. The highest rising edge frequencies of all the levels in the analysing frequency range are found. The invention imposes various conditions to find the most suitable rising edge frequency as the lowest frequency of the rightmost peak in the analysis frequency range. For instance, the number of level transitions from the sorted frequency to the processing end frequency should be continuously 2, the maximum number of level transitions should be 2, etc. The analysing frequency range is updated every time a new peak is found, and the analysis is repeated until the number of level transitions in all the levels in the analysing frequency range is zero. It is to be noted that a rising edge frequency is considered only if it is accompanied by a falling edge frequency and vice versa. After characterizing the spectrum, the target recognition is performed based on knowledge-based decision tree that is built using the observations of spectrums of different target classes under different conditions.
[0032] Figure 1 illustrates the first embodiment of the invention which describes various essential functional blocks involved in the proposed moving target classification method.
[0033] In one embodiment of the present invention depicted in Figure 1, there are four main functional blocks in the invention – Doppler Radar Receiver 1, Sampler 2, Spectrum Characterization 3 and Classifier 4. Doppler Radar transmits a signal which hits the targets in its field of view. The signal’s characteristics are changed based on the target it hits and is received back by the Radar. The echo signal received by the Radar is sampled for analysis. It is observed that the accuracy of the radar target classification also depends on the sampling frequency. Higher sampling frequency has better capability of capturing the target’s micro-motions also. The sampled signal is segmented and then fed into the spectrum characterization module which finds the details of the spectral peaks. Finally, the classifier recognizes the target based on the parameters obtained from the spectrum characterization module.
[0034] Figure 2 illustrates the second embodiment of the invention which describes a computationally less expensive, efficient and generic methodology to characterize the spectrum that can be used for any application where there is a need to know the characteristics of peaks in the spectrum.
[0035] In one embodiment of the present invention depicted in Figure 2, the Fast Fourier Transform of the input signal 5 whose spectrum needs to be characterized is computed 6. The absolute values of the computed FFT 7 are passed through a moving averaging filter 8. After filtering, the maximum and minimum amplitude values in the analyzing frequency range are computed 9. The amplitude range is quantized into N levels where the level width is calculated as (max amp value – min amp value)/N 10. The value of N is application dependent. Number of level transitions in the analyzing frequency range for all the levels are found and saved in the array OVERALL_LEVEL_TRANSITIONS[N] 11. Due to the presence of noise, it is possible that there are too many level transitions in the lower levels and there is no information in such levels. Therefore, the lowest processing level, the lowest level with number of level transitions less than N/2.5 is found from OVERALL_LEVEL_TRANSITIONS[N] for further analysis 12. The end_freq is initialized and the peak_count is initialized as 0 13. The initial end_freq should be set to a higher frequency if the application demands as the algorithm characterizes the spectrum from 0 Hz to the initial end_freq. The highest rising edge frequencies lesser than the end_freq corresponding to all the levels are computed 14, sorted and saved as sorted_freq[N]15. The processing starts with the highest sorted edge frequency 16, proceeds with the next highest frequency and so on 24, 25. The ref_highest_level, the possible highest level for any peak is set to N 17.
[0036] In an exemplary embodiment, the algorithm revolves around selecting the most suitable frequency among the sorted_freq[N] as the lowest frequency of the right most peak in the analyzing frequency range (i.e.), from 0 to end_freq. While selecting the most suitable lowest frequency of the peak, various conditions on level transitions from the sorted_freq to end_freq 18 are enforced (i) number of level transitions corresponding to all the levels from the highest level of the peak to the lowest level of the peak should be 2 and 0 in other levels 19 (ii) the maximum of number of level transitions should be 2 20 (iii) The computed highest_lvl 21 (the highest level with non-zero level transitions) should be lesser than or equal to ref_highest_level 22. The lowest frequency of the peak is found if one of the conditions mentioned above are not met. If all the conditions are met, the algorithm proceeds with the next highest sorted frequency after updating the ref_highest_level 23.
[0037] A peak has to exist at least in two levels in an exemplary embodiment. Therefore, after finding the lowest frequency of the peak, the computed LL & LH 26 are used for the validation of the peak 27. If a valid peak is found, various details of the peak such as lowest level (LL), highest level (LH), centre frequency (fC), lowest frequency (fL), highest frequency (fH), top frequency width of the peak and bottom frequency width of the peak are computed and saved in peaks_details[peak_count] 28 and the peak_count is incremented by 1 29. The end_freq keeps on decreasing as it is updated 30 with the lowest frequency of the recently found peak. Every time the end_freq is changed, OVERALL_LEVEL_TRANSITIONS[N] is re-computed 31. If the difference of the previous OVERALL_LEVEL_TRANSITIONS[N] and the re-computed OVERALL_LEVEL_TRANSITIONS[N] is not accounted for in the recently found peak details, the difference is accommodated in the peak after validation. The process of spectrum characterization ends 33 if all the values in the array OVERALL_LEVEL_TRANSITIONS[N] are zero 32. Otherwise, according to the updated end_freq, the process continues with the newly computed highest rising edge frequencies as explained above.
[0038] Once the details of all the peaks are found, merging of the peaks is done to avoid a single peak being detected as two peaks in an exemplary embodiment. For this, each peak’s lowest frequency, highest frequency and LH are re-estimated using the center frequency. Merging conditions are checked only if there is a difference in the LH and re-computed LH. If the neighboring detected peak has LH same as the peak being considered or the re-estimated LH and if it exists in lesser number of levels than the peak being considered, merging is done provided the neighbor’s center frequency lies between the re-estimated lowest frequency and the re-estimated highest frequency.
[0039] Figure 3 illustrates the third embodiment of the invention which describes an easily tunable target recognition scheme based on the inferences obtained from the spectrums of different classes under different conditions.
[0040] In an embodiment of the present invention depicted in Figure 3, the echo signal received by the Radar is the input 34 which needs to be classified as Moving Vehicle, Moving Pedestrian/Animal or Clutter. The input signal is fed to Spectrum Characterization module 35 which gives the required parameters of the spectrum. If the signal belongs to clutter, it is not necessary to check for the signature of a moving vehicle/pedestrian and therefore, classification of clutter is done first. Patterns corresponding to noise (identified using practical data) are classified as clutter 36, 37, 38, 39, 40, 41, 42, 43, 44, 45.
[0041] Spectrum of a moving vehicle exhibits proper peaks in the spectrum in an exemplary embodiment. Patterns corresponding to a moving vehicle are identified using practical data and are used for classification. Signal exhibiting a very strong peak is classified as a moving vehicle 46, 47, 48, 49, 50, 51, 52, 53, 54. Signal exhibiting one peak at a lower frequency covering more levels and another peak at a higher frequency covering at least a minimum number of levels is likely to belong to a slow moving vehicle and is classified as a moving vehicle 55, 56, 57, 58, 59, 60, 61, 62, 63. A moving vehicle exhibits a very small spectral width in the lower frequency band starting from 0 Hz. From the practical observations, the initial spectral width for moving vehicles is roughly less than a threshold WTh and for moving pedestrians, it is more than WTh. This initial width is computed 64. In order to classify a vehicle moving at a slower speed, if the initial width is less than WTh 65, the conditions for moving vehicle are relaxed and classification of moving vehicle is done 66, 67, 68, 69.
[0042] Further, once all the conditions for the classification of moving vehicle are checked, the signal is checked for moving pedestrian in an exemplary embodiment. Signals having a peak at higher frequencies should not be misclassified as moving pedestrian and such signals are classified as Clutter 70, 71, 72, 73. The signal is classified as moving pedestrian 77 if most of the samples are concentrated in the lower levels of the spectrum 74, 75 and the initial spectral width is greater than WTh 76. Otherwise, the signal is classified as Clutter.
[0043] In an embodiment of the present invention, a method for Pulsed Doppler Radar Target Recognition using a novel approach of spectrum characterization and knowledge-based decision tree is provided. The novel method of efficient and generic Spectrum Characterization technique computes the spectrum peak details such as the lowest level, highest level, lowest frequency, highest frequency, center frequency, top frequency width and bottom frequency width for all the peaks starting with the rightmost peak and ending with the leftmost peak of the spectrum. This characterization technique is not limited to Radar Target Recognition but could be used to any application where there is a need to characterize the spectrum.
[0044] In an exemplary embodiment, the quantization of levels is used for characterizing the spectrum. (i.e.), the filtered spectrum is quantized into N levels and number of level transitions (rising & falling edges) is computed for each level in the analyzing frequency range. The random transitions that arise due to noise at lower levels are eliminated by considering the levels higher than the lowest processing level only which is computed based on number of level transitions. A peak is found using the sorted rising edge frequencies in the processing frequency range and is considered valid if the number of level transitions from LL to LH between the processing sorted frequency and the processing end frequency is always two provided it exists at least in two levels.
[0045] Figure 4 illustrates a radar spectrum characterization system (400) in accordance with an embodiment of the present invention. The radar spectrum characterization system (400) includes a processor (402), a memory (404), an input unit (406), and an output unit (408).
[0046] The processor (402) receives the radar return signal. The processor(402) computes a Fast Fourier Transform (FFT) of the received radar return signal to obtain a spectrum. The processor(402) quantizes the spectrum into a plurality of amplitude levels. The processor(402) computes a lowest frequency (fL) of a rightmost peak of the quantized spectrum. The processor(402) determines a highest frequency (fH) of the rightmost peak of the quantized spectrum and a lowest quantized level (LL) and a highest quantized level (LH). The processor(402) computes a maximum amplitude (Amax), a minimum amplitude (Amin), a top frequency width, and a bottom frequency width. The processor(402) updates the lowest frequency (fL) as a processing end frequency after determining whether a peak is valid peak in the radar return signal.
[0047] The step of quantizing includes dividing the spectrum into the plurality of levels. Thereafter, the processor (402) determines a number of level transitions including rising and falling edges for each of the plurality of levels within a predefined frequency range.
[0048] The processor (402) sorts a plurality of rising edge frequencies in the predefined frequency range. Thereafter, the processor(402) determines the peak to be valid peak if the number of level transitions from the lowest quantized level (LL) to the highest quantized level (LH) between a processing sorted frequency and a processing end frequency is at least two.
[0049] The processor (402) re-estimates the lowest frequency (fL), the highest frequency (fH), and the highest quantized level (LH) based on the centre frequency of the peak. Thereafter the processor (402) merges the peak with adjacent peak when a centre frequency of the adjacent peak lies between the re-estimated lowest frequency (fL) and the re-estimated highest frequency (fH).
[0050] The radar return signal is classified as noise if a maximum absolute value of the spectrum is less than a predefined threshold (ATh).
[0051] The processor (402) determines a number of transitions in the spectrum. Thereafter, the processor (402) classifies the spectrum as noise if the number of transitions in the spectrum exceeds a predefined threshold number.
[0052] The radar return signal is classified as a moving vehicle if at least one of the following conditions is true: (i) the spectrum has at least one peak covering a plurality of levels and a centre frequency (fC) is greater than a predefined threshold frequency, (ii) the spectrum has at least two peaks having more levels than a predefined number of levels and corresponding centre frequencies (fC) are greater than the predefined threshold frequency, and (iii) the spectrum has a peak covering a plurality of levels and has a small initial spectrum width and a small peak frequency width.
[0053] The radar return signal is classified as a moving pedestrian if the spectrum has high initial spectrum width and does not have a peak at frequencies greater than a predefined threshold frequency.
[0054] The radar spectrum characterization method is executed for every valid peak of the radar return signal.
[0055] Advantageously, the problem of single peak getting detected as two peaks due to spurious peaks is addressed in a unique way. All the peaks’ lowest frequency, highest frequency and highest level are re-estimated using the information of the center frequency of the peaks. Merging is done if the neighboring peak’s center frequency lies between the re-estimated lowest frequency and the re-estimated highest frequency of the adjacent peak along with the other necessary conditions.
[0056] In yet another embodiment of the present invention, a new easily tunable approach of Target Recognition to classify the signal received by Radar to three different classes – Moving Pedestrian/Animal, Moving Vehicle and Clutter using knowledge-based decision tree is provided.
[0057] In an exemplary embodiment, the input data is classified as Clutter if the lowest processing level is too high & the average of level transitions from level 0 to the lowest processing level is greater than a fixed threshold. The input data is classified as Moving Vehicle using multiple conditions such as (a) if there is at least one peak covering a greater number of levels with center frequency greater than a predefined frequency (b) by checking for the presence of two peaks – one at a lower frequency covering more levels and the other at a comparatively higher frequency covering minimum number of levels and so on. The confusion of slow-moving vehicle and a moving pedestrian is handled by considering the presence of peak in the lower frequency range, initial spectrum width, top and bottom widths of the peak and the levels covered by the peak.
[0058] The foregoing description of the invention has been set merely to illustrate the invention and is not intended to be limiting. Since modifications of the disclosed embodiments incorporating the spirit and substance of the invention may occur to person skilled in the art, the invention should be construed to include everything within the scope of the invention.
,CLAIMS:
1. A radar spectrum characterization method for determining a valid peak in a radar return signal, the method comprising:
receiving, by a processor, the radar return signal;
computing, by the processor, Fast Fourier Transform (FFT) of the received radar return signal to obtain a spectrum;
quantizing, by the processor, the spectrum into a plurality of amplitude levels;
computing, by the processor, lowest frequency (fL) of a rightmost peak of the quantized spectrum;
determining, by the processor, highest frequency (fH) of the rightmost peak of the quantized spectrum and lowest quantized level (LL) and highest quantized level (LH);
computing, by the processor, maximum amplitude (Amax), minimum amplitude (Amin), top frequency width, and bottom frequency width; and
updating, by the processor, the lowest frequency (fL) as processing end frequency after determining whether a peak is valid peak in the radar return signal.

2. The radar spectrum characterization method as claimed in claim 1, wherein the step of quantizing includes:
dividing, by the processor, the spectrum into the plurality of levels; and
computing, by the processor, a number of level transitions including rising and falling edges for each of the plurality of levels within a predefined frequency range.

3. The radar spectrum characterization method as claimed in claim 2, further comprising:
sorting, by the processor, a plurality of rising edge frequencies in the predefined frequency range; and
determining, by the processor, the peak to be valid peak if the number of level transitions from the lowest quantized level (LL) to the highest quantized level (LH) between processing sorted frequency and processing end frequency is at least two.

4. The radar spectrum characterization method as claimed in claim 1, further comprising:
re-estimating, by the processor, the lowest frequency (fL), the highest frequency (fH), and the highest quantized level (LH) based on the centre frequency of the peak; and
merging, by the processor, the peak with adjacent peak when a centre frequency of the adjacent peak lies between the re-estimated lowest frequency (fL) and the re-estimated highest frequency (fH).

5. The radar spectrum characterization method as claimed in claim 1, wherein the radar return signal is classified as noise if a maximum absolute value of the spectrum is less than a predefined threshold (ATh).

6. The radar spectrum characterization method as claimed in claim 1, further comprising:
determining, by the processor, a number of transitions in the spectrum; and
classifying, by the processor, the spectrum as noise if the number of transitions in the spectrum exceeds a predefined threshold number.

7. The radar spectrum characterization method as claimed in claim 1, wherein the radar return signal is classified as a moving vehicle if at least one of the following conditions is true:
the spectrum has at least one peak covering a plurality of levels and a centre frequency (fC) is greater than a predefined threshold frequency,
the spectrum has at least two peaks having more levels than a predefined number of levels and corresponding centre frequencies (fC) are greater than the predefined threshold frequency, and
the spectrum has a peak covering a plurality of levels and has a small initial spectrum width and a small peak frequency width.

8. The radar spectrum characterization method as claimed in claim 1, wherein the radar return signal is classified as a moving pedestrian if the spectrum has high initial spectrum width and does not have a peak at frequencies greater than a predefined threshold frequency.

9. The radar spectrum characterization method as claimed in claim 1, wherein said radar spectrum characterization method is executed for every valid peak of the radar return signal.