def push_back(self,I):
        """
        Compute features for current image, push to stack

        """
        h,w = I.shape[:2]
        self.h = h
        self.w = w

        if self.params['features_clahe'] > 0:
            Ibw = clahe.convert_bw(I)
        else:
            if I.ndim > 2:
                Ibw = cv2.cvtColor(I,7)
            else:
                Ibw = I

        for i,s in enumerate(self.scale_ranges):
            hn = int(np.round(h/s))
            wn = int(np.round(w/s))
            
            # WORKAROUNDS for some quirks in libviso2.
            # We need to make sure that the image has an odd number of rows, and an
            # even number of columns.
            hn = hn - (hn % 2 == 0)
            wn = wn - (wn % 2 == 1)

            I_ = cv2.resize(Ibw,(wn,hn))

            self.matchers[i].pushBack(I_)
Пример #2
0
    def push_back(self, I):
        """
        Compute features for current image, push to stack

        """
        h, w = I.shape[:2]
        self.h = h
        self.w = w

        if self.params['features_clahe'] > 0:
            Ibw = clahe.convert_bw(I)
        else:
            if I.ndim > 2:
                Ibw = cv2.cvtColor(I, 7)
            else:
                Ibw = I

        for i, s in enumerate(self.scale_ranges):
            hn = int(np.round(h / s))
            wn = int(np.round(w / s))

            # WORKAROUNDS for some quirks in libviso2.
            # We need to make sure that the image has an odd number of rows, and an
            # even number of columns.
            hn = hn - (hn % 2 == 0)
            wn = wn - (wn % 2 == 1)

            I_ = cv2.resize(Ibw, (wn, hn))

            self.matchers[i].pushBack(I_)
Пример #3
0
    def push_back(self, I):
        """
        Compute features for current image, push to stack

        """
        h, w = I.shape[:2]
        self.h = h
        self.w = w

        if self.params['features_clahe'] > 0:
            if I.ndim > 2:
                I_ = clahe.convert_color(I)
            else:
                I_ = clahe.convert_bw(I)
        else:
            I_ = I.copy()

        print(I_.dtype)

        kp_in_opencv, desc = self.extractor.detectAndCompute(I_, None)
        kp_in = np.array([P.pt for P in kp_in_opencv])

        if self.params['features_prune_border'] > 0:
            prune = self.params['features_prune_border']
            border_x = w * prune
            border_y = h * prune

            ind_valid = np.logical_and(
                np.logical_and(kp_in[:, 0] >= border_x,
                               kp_in[:, 0] <= w - border_x),
                np.logical_and(kp_in[:, 1] >= border_y,
                               kp_in[:, 1] <= h - border_y))
            kp_in = kp_in[ind_valid, :]

            # Clunky filtering of python list
            # (Doing this via list(array(...)[inds]) is much slower).
            kp_in_opencv = [
                kp_in_opencv[i] for i in xrange(len(kp_in_opencv))
                if ind_valid[i]
            ]
            desc = desc[ind_valid, :]

        #kp,desc = self.brisk_descriptor_extractor.compute(I_,kp_in_opencv)

        #if I_.ndim > 2:
        #    kp,desc = self.opponentbrief_descriptor_extractor.compute(I_,kp_in)
        #else:
        #    kp,desc = self.brief_descriptor_extractor.compute(I_,kp_in)

        self.keypoints.append(kp_in_opencv)
        self.descriptors.append(desc)
    def push_back(self,I):
        """
        Compute features for current image, push to stack

        """
        h,w = I.shape[:2]
        self.h = h
        self.w = w

        if self.params['features_clahe'] > 0:
            if I.ndim > 2:
                I_ = clahe.convert_color(I)
            else:
                I_ = clahe.convert_bw(I)
        else:
            I_ = I.copy()

        print(I_.dtype)

        kp_in_opencv,desc = self.extractor.detectAndCompute(I_,None)
        kp_in = np.array([P.pt for P in kp_in_opencv])

        if self.params['features_prune_border'] > 0:
            prune = self.params['features_prune_border']
            border_x = w * prune
            border_y = h * prune

            ind_valid = np.logical_and(np.logical_and(kp_in[:,0] >= border_x,
                                                      kp_in[:,0] <= w-border_x),
                                       np.logical_and(kp_in[:,1] >= border_y,
                                                      kp_in[:,1] <= h-border_y))
            kp_in = kp_in[ind_valid,:]

            # Clunky filtering of python list
            # (Doing this via list(array(...)[inds]) is much slower).
            kp_in_opencv = [kp_in_opencv[i] for i in xrange(len(kp_in_opencv)) if ind_valid[i]]
            desc = desc[ind_valid,:]

        
        #kp,desc = self.brisk_descriptor_extractor.compute(I_,kp_in_opencv)

        #if I_.ndim > 2:
        #    kp,desc = self.opponentbrief_descriptor_extractor.compute(I_,kp_in)
        #else:
        #    kp,desc = self.brief_descriptor_extractor.compute(I_,kp_in)
       
        self.keypoints.append(kp_in_opencv)
        self.descriptors.append(desc)