Cluster2FindAnomalies.py

# coding: utf-8

# In[155]:

import matplotlib
matplotlib.use('Agg')
import math
import matplotlib.pyplot as plt
import numpy
import scipy.cluster.hierarchy as hcluster
from sklearn.preprocessing import StandardScaler
import pandas as pd
import glob
import os
import numpy as np
import cv2
import sys

vid_name = sys.argv[1]


# In[182]:

# Velocity Trace - Fast Cars

for f in glob.glob('projections/' + vid_name + '_*.csv'):
    # Get Video ID
    vid_id =  f.split("projections/")[1].split("_")[0]
    # Read in Data
    data = pd.read_csv(f, header=None)
    if "fast_cars" in f:
        continue
        data.columns = ["Vid", "Object", "C"]
        cluster_var = ['C']
        save_name = "_fast_cars"
        data_group_name = "_withVelocity.csv"
        data_cols = ['cog_x','cog_y','frame','height','obj','velocity','width','x','y']
    elif "slow_cars" in f:
        continue
        data.columns = ["Vid", "Object", "C"]
        cluster_var = ['C']
        save_name = "_slow_cars"
        data_group_name = "_withVelocity.csv"
        data_cols = ['cog_x','cog_y','frame','height','obj','velocity','width','x','y']
    elif "maintain_velocity" in f:
        data.columns = ["Vid", "Object", "C1", "C2"]
        cluster_var = ['C1', 'C2']
        save_name = "_maintain_velocity"
        data_group_name = "_withVelocity.csv"
        data_cols = ['cog_x','cog_y','frame','height','obj','velocity','width','x','y']

        avg_c1 = data['C1'].mean()
        avg_c2 = data['C2'].mean()
        data = data[(data['C1'] > 30) & (data['C2'] < 30)]
    elif "to_shoulder" in f or "to_lane" in f:
        continue
        data.columns = ["Vid", "Object", "w", "tau"]
        cluster_var = ['w', 'tau']
        save_name = "_to_shoulder" if "to_shoulder" in f else "to_lane"
        data_group_name = "_withLanes.csv"
        data_cols = ['height', 'width', 'y','x','cog_y','cog_x', 'obj', 'frame', 'lane']
    elif "in_lane" in f:
	continue
        data.columns = ["Vid", "Object", "tau"]
        cluster_var = ['tau']
        save_name = "_in_lane"
        data_group_name = "_withLanes.csv"
        data_cols = ['height', 'width', 'y','x','cog_y','cog_x', 'obj', 'frame', 'lane']
    else:
        continue
    # Cluster
    X = np.asarray(data[cluster_var]).reshape((len(data[cluster_var[0]]), len(cluster_var)))
    clusters = hcluster.fclusterdata(X, 1.5, criterion="distance")
    # Save Cluster Plot and Cluster Results
    data_results = data
    data_results['Cluster'] = clusters 
    if len(cluster_var) == 1:
        if save_name == "_in_lane":
            plt.scatter(range(0, len(data['tau'])) ,data['tau'].values, c=clusters)
            plt.xlabel('t')
            plt.ylabel('tau')
        else:
            plt.scatter(range(0, len(data['C'])) ,data['C'].values, c=clusters)
            plt.xlabel('t')
            plt.ylabel('C')
    else:
        if data_group_name == "_withVelocity.csv":
            plt.scatter(data['C1'].values, data['C2'].values, c=clusters)
            plt.xlabel('C1')
            plt.ylabel('C2')
        else:
            plt.scatter(data['w'].values, data['tau'].values, c=clusters)
            plt.xlabel('w')
            plt.ylabel('tau')
    plt.savefig("cluster_plots/" + vid_id + save_name + ".png")
    # Find Smallest Cluster
    data_group = data[['Object', 'Cluster']].groupby(['Cluster']).agg(['count'])
    data_group['Cluster'] = data_group.index

    # Find Smallest Cluster if Velocity
    if data_group_name == "_withVelocity.csv" and save_name != "_maintain_velocity":
        smallest_cluster = [data_group.sort_values([('Object', 'count')])['Cluster'].iloc[0]]
        # Get Object IDs
        anomaly_objs = data['Object'][data['Cluster'].isin(smallest_cluster)]
    elif save_name == "_in_lane":
        # Get Object IDs
#        anomaly_objs = data['Object'][data['tau'] > 0]
        avg_tau = data['tau'].mean()
        avg_std = data['tau'].std()
        anomaly_objs = data['Object'][data['tau'] > (avg_tau + avg_std/2.0)]
    elif save_name == "_maintain_velocity":
#        avg_c1 = data['C1'].mean()
#        avg_c2 = data['C2'].mean()
#        data = data[(data['C1'] > avg_c1) & data['C2'] > avg_c2]
        data_group = data[['Object', 'Cluster']].groupby(['Cluster']).agg(['count'])
        data_group['Cluster'] = data_group.index
        avg_cluster_size = data_group[('Object', 'count')].mean()
        std_cluster_size = data_group[('Object', 'count')].std()
        smallest_cluster = data_group[data_group[('Object', 'count')] < math.ceil(avg_cluster_size/2.0)]['Cluster'].values
        # Get Object IDs
        anomaly_objs = data['Object'][data['Cluster'].isin(smallest_cluster)]
    else:
        # Find Smaller than Average Clusters if Lane
        avg_cluster_size = data_group[('Object', 'count')].mean()
        std_cluster_size = data_group[('Object', 'count')].std()
        smallest_cluster = data_group[data_group[('Object', 'count')] < abs(avg_cluster_size - (std_cluster_size/2.0))]['Cluster'].values
        # Get Object IDs
        anomaly_objs = data['Object'][data['Cluster'].isin(smallest_cluster)]
    # Get X, Y, Height, Length, and Frame IDs
    anomaly_obj_data = pd.read_csv("data/" + str(vid_id) + data_group_name, header=None)
    anomaly_obj_data.columns = data_cols
    anomaly_obj_data = anomaly_obj_data[['obj', 'x', 'y', 'height', 'width', 'frame']]
    anomaly_obj_data = anomaly_obj_data[anomaly_obj_data['obj'].isin(anomaly_objs)]
    # Save File with Object ID, X, Y, Height, Length, Frame IDs
    anomaly_obj_data.to_csv("anomalies/" + str(vid_id) + save_name + ".csv")
    # Draws Bounding Boxes
    frames = anomaly_obj_data['frame'].unique()
    for i in frames: 
        image_name = str(vid_id) + "/" + str(i) + '.png'
	print image_name
        img = cv2.imread(image_name)
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        anomaly_cars_in_frame = anomaly_obj_data[anomaly_obj_data['frame'] == int(i)]
        for index, row in anomaly_cars_in_frame.iterrows():
            x1 = row['x']
            y1 = row['y']
            h = row['height']
            w = row['width']
            cv2.rectangle(img, (int(x1), int(y1)), (int(x1 + w), int(y1 + h)), (0, 0, 255), 3)
        labeled_image_name = "anomaly_pics_mv/" + vid_id + "_" + str(i) + save_name + ".png"
        cv2.imwrite(labeled_image_name, img)
    



# In[ ]: