Method for detecting large size and passenger vehicles from fixed cameras

• US 9,317,752 B2
• Filed: 06/20/2013
• Issued: 04/19/2016
• Est. Priority Date: 06/20/2013
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Bulan, Orhan, Wu, Wencheng, Wang, Yao Rong, Loce, Robert P.
• Primary Examiner(s): Park, Chan
• Assistant Examiner(s): Summers, Geoffrey E
• Application Number:
  US13/922,336
• Publication Number:
  US 20140376769A1
• Time in Patent Office: 1,034 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69

Abstract

A method for detecting parking occupancy includes receiving video data from a sequence of frames taken from an associated image capture device monitoring a parking area. The method includes determining at least one candidate region in the parking area. The method includes comparing a size of the candidate region to a size threshold. In response to size of the candidate region meeting and exceeding the size threshold, the method includes determining whether the candidate region includes one of at least one object and no objects. The method includes classifying at least one object in the candidate region as belonging to one of at least two vehicle-types. The method further includes providing vehicle occupancy information to a user.

Claims

1. A method for detecting a vehicle, the method comprising:

receiving video data from a sequence of frames taken from an associated image capture device monitoring a parking area;

in response to detecting an object in a current frame, associating the detected object as a candidate region where a newly parked vehicle may exist;

in response to a size of the candidate region meeting or exceeding a size threshold, performing a sliding window search limited to the candidate region of each frame to extract object features of the detected candidate region from each of multiple windows;

applying the extracted object features of each window to a first type vehicle classifier trained to classify larger vehicles that are divided into multiple parts each corresponding to a different window in the sliding window search;

acquiring a first score from the first vehicle classifier indicative of a degree that the extracted object features match a trained vehicle of the first vehicle classifier;

in response to the first score meeting or exceeding a first confidence threshold, associating the newly parked vehicle with a large vehicle of a size that extends beyond the bounds of a single parking space;

in response to the first score not meeting or exceeding the first confidence threshold, ruling out the large vehicle and associating the detected object as a candidate region where one or more of a small vehicle and shadow may exist;

also in response to the first score not meeting or exceeding the first confidence threshold, for each window, applying the extracted object features to a second type vehicle classifier trained to classify vehicles that are smaller-sized than the large vehicle;

acquiring a second score from the second type vehicle classifier indicative of a degree that the extracted object features match a trained vehicle of the second vehicle type;

in response to the second score meeting or exceeding a second confidence threshold indicative of a degree that the extracted object features match a trained second vehicle of the second vehicle classifier, associating the newly parked vehicle with one or more small vehicles;

in response to the second score not meeting or exceeding the second confidence threshold, associating the candidate region as not containing a vehicle; and

providing information regarding an occupancy of the parking space to a user device base on the associations.

2. The method of claim 1, wherein the detecting an object in a current frame includes:

tracking a vehicle into a parking spot using a coherent cluster of motion vectors whose length go to zero in the parking area; and

,

identifying a collection of pixels that were last associated with the coherent cluster of motion vectors.

3. The method of claim 1, wherein the size is calculated by one of calculating a physical length of the candidate region using real-world coordinates mapped from pixel coordinates and calculating an area of the candidate region on the image plane.

4. The method of claim 1

wherein the user device is associated with a user of vehicles located in the parking area.

5. A computer program product comprising a non-transitory computer readable memory storing instructions for performing a method of:

receiving video data from a sequence of frames taken from an associated image capture device monitoring a parking area;

in response to detecting an object in a current frame, associating the detected object as a candidate region where a newly parked vehicle may exist;

in response to a size of the candidate region meeting or exceeding a size threshold, performing a sliding window search limited to the candidate region off each frame to extract object features of the detected candidate region from each of multiple windows;

applying the extracted object features of each window to a first type vehicle classifier trained to classify larger vehicles that are a size capable of being divided into multiple parts each being in a different window of the sliding window searches;

acquiring a first score from the first vehicle classifier indicative of a degree that the extracted object features match a trained vehicle of the first vehicle classifier;

in response to the first score not meeting or exceeding the first confidence threshold, associating the newly parked vehicle with a large vehicle of a size that extends beyond the bounds of a single parking space;

in response to the score not meeting the confidence threshold, ruling out the larger vehicle and associating the detected object as a candidate region where one or more of a small vehicle and shadow may exist;

also in response to the first score not meeting or exceeding the first confidence threshold, for each window, applying the extracted object features to a second type vehicle classifier trained to classify vehicles that are smaller-sized than the large vehicle;

acquiring a second score from the second type vehicle classifier indicative of a degree that the extracted object features match a trained vehicle of the second vehicle type;

in response to the second score meeting or exceeding a second confidence threshold indicative of a degree that the object features match a trained second vehicle of the second vehicle classifier, associating the newly parked vehicle with one or more small vehicles;

in response to the second score not meeting or exceeding the second confidence threshold, associating the candidate region as not containing a vehicle; and

providing information regarding an occupancy of the parking space to a user device base on the associations.

6. A monitoring system for detecting a vehicle, the system comprising a vehicle detection device including a processor configured to:

receive video data from a sequence of frames taken from an associated image capture device monitoring a parking area;

in response to detecting an object in a current frame, associate the detected object as a candidate region where a newly parked vehicle may exist;

in response to a size of the candidate region meeting or exceeding a size threshold, perform a sliding window search limited to the candidate region of each frame to extract object features of the detected candidate region from each of multiple windows;

apply the extracted object features of each window to a first type vehicle classifier trained to classify larger vehicles that can be divided into multiple parts each in a different window of the sliding window search;

acquire a first score from the first vehicle classifier indicative of a degree that the extracted object features match a trained vehicle of the first vehicle classifier;

in response to the first score meeting or exceeding a first confidence threshold, associate the newly parked vehicle with a large vehicle of a size that extends beyond the bounds of a single parking space;

in response to the first score not meeting or exceeding the first confidence threshold, rule out the large vehicle and associate the detected object as a candidate region where one or more of a small vehicle and shadow may exist;

also in response to the first score not meeting or exceeding the first confidence threshold, for each window, apply the object features to a second type vehicle classifier trained to classify vehicles that are smaller size than the large vehicle;

acquire a second score from the second type vehicle classifier indicative of a degree that the extracted object features match a trained vehicle of the second vehicle type;

in response to the second score meeting or exceeding a second confidence threshold indicative of a degree that the object features match a trained second vehicle of the second vehicle classifier, associate the detected object with one or more small vehicles occupying the single one or more parking spaces;

in response to the second score not meeting or exceeding the second confidence threshold, associate the parking space as not containing a vehicle; and

provide information regarding an occupancy of the parking space to a user device base on the associations.

7. The system of claim 6, wherein the processor is further adapted to:

classify the object as one of a motorcycle, a passenger vehicle and a commercial vehicle.

8. The system of claim 6, wherein the processor is further adapted to:

track a vehicle into a parking spot using a coherent cluster of motion vectors whose length go to zero in the parking area; and

,

identify a collection of pixels that were last associated with the coherent cluster of motion vectors.

9. The system of claim 8, wherein the processor is further adapted to determine the size of the candidate region by:

extracting at least one feature from the object using a binary mask;

calculate the size using the at least one extracted feature.

10. The system of claim 6, wherein the processor is adapted to calculate the size by one of calculating a physical length of the candidate region using real-world coordinates mapped from pixel coordinates and calculating an area of the candidate region on the image plane.